Embed Size (px)
Citation preview
1
ASSESSMENT OF COMMERCIAL IMAGE PROCESSING SOFTWARE PROGRAMS FOR UNMANNED AUTONOMOUS VEHICLE IMAGERY
By
HECTOR YAMIL RODRIGUEZ ASILIS
A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF MASTER IN SCIENCE
UNIVERSITY OF FLORIDA
2012
2
© 2012 Héctor Yamil Rodríguez Asilis
3
To Abuela Ticha
4
ACKNOWLEDGMENTS
To my Dad and Mom, for supporting me in every decision I have made from the
moment I was born, all the way through school and later my undergraduate studies, and
also in the graduate studies I am completing now.
To my girlfriend Olga and my sister Yamilé, the two women of my life, they
understand me in the good and the hard times.
To all the wonderful people I met in Gainesville, which every one of them have
become family, they have been by my side in the great times and in the bad times in my
journey through this town, with my beloved ones more than a thousand miles away, I
will miss all of you, Ronald, Nico, Rodolfo, Gonzalo, Leonardo, Oscar, Pedro,
Christobal, Robert, Kurtis, Luis, Julio, Kimmel, Bora, Alessio, Felix, Ivelisse, Angelica,
Gisselle, Sonia, Rocio, Silvia, Sofia, Martha, Angie, Luisa.
To my Advisor Dr. Scot Smith, for being my support during this wonderful journey
in the world of the Geomatics, in a totally new country for me.
To the University of Florida Unmanned Aerial Systems Research Group
(UFUASRG) and the U.S. Army Corps of Engineers for giving me the opportunity to
work with them in the post-processing and for trusting me, Matt Burgess, Bon Dewitt,
Peter Ifju, Frank Percival, , Scot Smith, Damon Wolfe, John Perry, Tyler Ward and
others.
5
TABLE OF CONTENTS page
ACKNOWLEDGMENTS .................................................................................................. 4
LIST OF TABLES ............................................................................................................ 8
LIST OF FIGURES .......................................................................................................... 9
LIST OF ABBREVIATIONS ........................................................................................... 12
ABSTRACT ................................................................................................................... 14
CHAPTER
1 INTRODUCTION .................................................................................................... 16
Unmanned Aerial Systems Research Project ......................................................... 16 Unmanned Aerial Systems ..................................................................................... 18
Limitations of UAVs................................................................................................. 19 Photogrammetry ..................................................................................................... 19
Aerial Photogrammetry ..................................................................................... 20 UAS Photogrammetry....................................................................................... 21
Comparison ...................................................................................................... 21 Automatic Tie-point Extraction .......................................................................... 22
Aerial Triangulation Techniques ....................................................................... 24 Bundle Block Adjustment ........................................................................... 25
Use of GPS in Aerotriangulation ................................................................ 26 Determination of the Attitude of the Aircraft ............................................... 27
2 LITERATURE REVIEW .......................................................................................... 29
UAS Uses and Applications .............................................................................. 29
Advances in Automatic Tie Point Extraction and Aerial Triangulation .............. 30
3 METHODS .............................................................................................................. 34
Study Area .............................................................................................................. 34 Determination of Ground Elevation on Center of Images........................................ 35
Calculation of Overlap Percentages........................................................................ 36 Tree and Cloud Coverage per Image...................................................................... 37
Workstation Specifications ...................................................................................... 37 UF UAS Payload Output Data ................................................................................ 37
Non-Vegetated Area of Study ................................................................................. 38
4 RESULTS AND DISCUSSION ............................................................................... 44
6
2d3 Altimap ............................................................................................................. 44 2d3 Sensing ..................................................................................................... 45
2d3 Data Preparation ....................................................................................... 46 Image Cleaning ................................................................................................ 46
Block Creation .................................................................................................. 46 Total Processing Times.............................................................................. 48
Processing Time Results ........................................................................... 48 Maximum Quantity of Images Processed by 2d3 ....................................... 50
Import Exif Data to Images ............................................................................... 50 Creating Image Directory.................................................................................. 51
Processing ........................................................................................................ 51 Processing Report ............................................................................................ 52
Processing Time Assessment .......................................................................... 52 Overlap Percentage Assessment ..................................................................... 53
Cloud and Tree Impact Assessment ................................................................ 53 EnsoMOSAIC ......................................................................................................... 54
Mosaic Mill ........................................................................................................ 55 Input Data ......................................................................................................... 55
Creation of Blocks ............................................................................................ 56 TRP File ..................................................................................................... 57
GPS File..................................................................................................... 57 CAL File ..................................................................................................... 58
Pyramid Images ............................................................................................... 58 Automatic v7 Aerial Triangulation ..................................................................... 58
Ortho-rectification ............................................................................................. 60 Erdas LPS ............................................................................................................... 61
Erdas ImageStation .......................................................................................... 62 Erdas ................................................................................................................ 62
Image Cleaning and Input Files ........................................................................ 62 Perform Automatic Tie Point Generation .......................................................... 63
Tie Point Generation Experiment ............................................................... 64 Auto Tie Point Generation Assessment ..................................................... 65
Aerial Triangulation .......................................................................................... 66 Ortho-rectification ............................................................................................. 68
DTM Extraction .......................................................................................... 68 Ortho-Resampling ...................................................................................... 68
Mosaic Creation ......................................................................................... 69
5 CONCLUSIONS ..................................................................................................... 86
Comparison ............................................................................................................ 86 Conclusions ............................................................................................................ 87
Recommendations .................................................................................................. 88
APPENDIX IMAGE EXIF DATA READERS .............................................................. 93
LIST OF REFERENCES ............................................................................................... 95
7
BIOGRAPHICAL SKETCH ............................................................................................ 99
8
LIST OF TABLES
Table page 3-1 Block Configuration and Properties .................................................................... 39
3-2 Block Configuration Flight Lines and Directions ................................................. 39
4-1 Sea Horse Key, Cedar Key Data Set Block Configuration. ................................. 70
4-2 Total Processing Time Per Configuration ........................................................... 71
4-3 Processing Results ............................................................................................. 71
4-4 Configuration and Options Used for ATP Experiment ........................................ 72
9
LIST OF FIGURES
Figure page 1-1 Principle of bundle-block adjustment.. ................................................................ 28
3-1 Map of flight mission.. ......................................................................................... 40
3-2 Contour Line Data, Flight Plan and Digital Elevation Model.. ............................. 40
3-3 Image Exterior Orientation Parameters: Omega, Phi and Kappa. ...................... 41
3-4 Non-vegetated area of study. ............................................................................. 42
3-5 Block Configurations. .......................................................................................... 43
4-1 Sea Horse Key Data Set. ................................................................................... 72
4-2 Scaling error after mosaicking a big number of images. ..................................... 73
4-3 Images Processed vs. Time Processed Chart.. .................................................. 73
4-4 Images Processed vs. Time Processed Chart for Block Configuration 3.. .......... 74
4-5 Chart showing the relations between average overlap to processing time.. ....... 74
4-6 Relation between final mosaic area with processing time per block.. ................. 75
4-7 Relationship between number of images utilized in the final mosaicked product and the processing time.. ....................................................................... 75
4-8 Relationship between Processing Time and Total Images in Block.. .................. 76
4-9 Relation between Image by Processing Minute and Forward Overlap.. ............. 76
4-10 Scatter Plot Showing Relation Between Image Pairs Matched and Forward Overlap.. ............................................................................................................. 77
4-11 Scatter Plot of Relation Between Images Mosaicked and Forward Overlap.. ..... 77
4-12 Scatter plot showing relation image percent with cloud and tree present.. ......... 78
4-13 RMSE variation through measurements stages in Aerial Automatic Triangulation.. ..................................................................................................... 78
4-14 Scatter plot showing relation between overlap percentages with quantity of tie points found on the initial stage. .................................................................... 79
10
4-15 Scatter plot showing relation between total images in the block with quantity of tie points found on the initial stage.. ............................................................... 79
4-16 Scatter plot showing relation between overlap percentages in the block with accuracy of tie points found on the initial stage. ................................................. 80
4-17 Cloud presence and high pitch-roll values impact on tie point search. ............... 80
4-18 Scatter plot showing relation between overlap percentages in the block with Root Mean Square Errors on the Aerial Triangulation process........................... 81
4-19 Scatter plot showing relation Root Mean Square Errors on the Aerial Triangulation process and percent of images used for mosaicking.. .................. 81
4-20 ATP Experiment Accuracy per Configuration. .................................................... 82
4-21 Scatter Plot Showing Percent Right vs. ATP Found. .......................................... 82
4-22 Scatter plot showing relation between tie point accuracy and overlap percentages. ....................................................................................................... 83
4-23 Scatter plot showing relation between overlap percentages with quantity of tie points found on the initial stage. .................................................................... 83
4-24 Scatter plot showing relation between total images in the block with quantity of tie points found on the initial stage. ................................................................ 84
4-25 Cloud presence and high pitch-roll values impact on tie point search. ............... 84
4-26 Relation between overlap percentages and Root Mean Square Error (RMSE) in aerotriangulation. ............................................................................................ 85
5-1 Resulting mosaic using 2d3 Altimap ................................................................... 89
5-2 Resulting mosaic using EnsoMOSAIC. .............................................................. 90
5-3 Resulting mosaic using Erdas LPS.. ................................................................... 90
5-4 Comparison between EnsoMOSAIC and Erdas LPS in automatic tie point found. ................................................................................................................. 91
5-5 Comparison between EnsoMOSAIC and Erdas LPS in automatic tie point accuracy. ............................................................................................................ 91
5-6 Comparison between EnsoMOSAIC and Erdas LPS in aerial triangulation error. ................................................................................................................... 92
A-1 Jeffrey´s Exif Viewer, showing the location of an Archer Field Mission Image using the EXIF info written to it. .......................................................................... 93
11
A-2 Example of all the data contained in EXIF format written in an image.. .............. 94
12
LIST OF ABBREVIATIONS
2D TWO DIMENSIONAL
3D THREE DIMENSIONAL
AP AERIAL PHOTOGRAMMETRY
AAT AUTOMATIC AERIAL TRIANGULATION
ATP AUTOMATIC TIE POINT
CE CIVIL ENGINEERING
CMD COMMAND PROMPT
DEM DIGITAL ELEVATION MODEL
DGRS DIRECTLY GEOREFERENCED REMOTE SENSING
DSM DIGITAL SURFACE MODEL
DTM DIGITAL TERRAIN MODEL
EM ENSOMOSAIC
EOP EXTERIOR ORIENTATION PARAMETER
EXIF EXCHANGEABLE IMAGE FILE FORMAT
FAA FEDERAL AVIATION ADMINISTRATION
FGDL FLORIDA GEOGRAPHICAL DATA LIBRARIES
FWMD FLORIDA WATER MANAGEMENT DISTRICTS
FWS FLORIDA FISH AND WILDLIFE SERVICE
GCP GROUND CONTROL POINTS
GNSS GLOBAL NAVIGATION SATELLITE SYSTEM
GPS GLOBAL POSITIONING SYSTEM
GSD GROUND SAMPLE DISTANCE
IMU INERTIAL MEASUREMENT UNIT
INS INERTIAL NAVIGATION SYSTEMS
13
IOP INTERIOR ORIENTATION PARAMETERS
ISO INTERNATIONAL ORGANIZATION FOR STANDARDIZATION
KML KEYHOLE MARKUP LANGUAGE
KPO KAPPA PHI OMEGA
LPS LEICA PHOTOGRAMMETRY SUITE
NGS NATIONAL GEODETIC SURVEY
OTF ON THE FLY
RMSE ROOT MEAN SQUARE ERROR
RPY ROLL PITCH YAW
RTK REAL TIME KINEMATIC
SLR SINGLE LENS REFLEX
SUAS SMALL UNMANNED AERIAL SYSTEMS
SUAV SMALL UNMANNED AUTONOMOUS VEHICLE
UAS UNMANNED AERIAL SYSTEM
UAV UNMANNED AUTONOMOUS VEHICLE
UAVP UAV PHOTOGRAMMETRY
UF UNIVERSITY OF FLORIDA
UFUASRG UNIVERSITY OF FLORIDA UNMANNED AERIAL SYSTEMS RESEARCH
GROUP
USACE UNITED STATES ARMY CORPS OF ENGINEERS
USGS UNITED STATES GEOLOGICAL SURVEY
UTC COORDINATED UNIVERSAL TIME
UTM UNIVERSAL TRANSVERSE MERCATOR
VTOL VERTICAL TAKE-OFF AND LANDING
WGS84 WORLD GEODETIC SYSTEM 1984
14
Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science
ASSESSMENT OF COMMERCIAL IMAGE PROCESSING SOFTWARE PROGRAMS FOR UNMANNED AUTONOMOUS VEHICLE IMAGERY
By
Héctor Yamil Rodríguez Asilis
December 2012
Chair: Scot Smith Major: Forest Resources and Conservation
An Unmanned Aircraft System (UAS) was used to acquire digital images and
produce geo-rectified mosaics, providing data and images to support ecosystem
restoration, invasive species control monitoring, levee safety monitoring, and
emergency natural disaster response.
A weakness of the system is extensive data post-processing requirement. The
large volume of data collected on typical missions makes automated processing
attractive. One of the primary advantages for the use of this system was the speed and
low cost with which it can be deployed, but delays introduced by current data
processing workflows that require extensive manual effort reduce this capability.
The research in this thesis was to investigate, (1) evaluate and identify appropriate
software to streamline the pre/post-processing for delivering geospatial data in support
of the topics described above, (2)enhance the processing workflow which includes
camera geometric and radiometric calibration, (3)radiometric data processing,
geospatial data pre-processing, (4)sparse tie point generation, (5)photogrammetric
adjustment, (6)tie point densification, (7)terrain generation, (8)seam line generation,
(9)radiometric correction, (10)mosaic generation and (11)establish an appropriate
15
amount of side slap coverage and overlap area for the proper flight appropriate needed
according to the software being evaluated.
Three software programs were evaluated: (1) Erdas LPS, (2) 2d3 Altimap and (3)
EnsoMosaic: The expected results were that: (a) the 2d3 Altimap would be the fastest
and require less manual input, but that it would produce a less accurate solution, (b)
EnsoMosaic would be the most accurate of the three and require less manual input than
Erdas LPS, but be relatively slow.
16
CHAPTER 1 INTRODUCTION
Unmanned Aerial Systems Research Project
The Unmanned Aerial Systems Research Group, Florida Cooperative Fish and
Wildlife Research Unit's multidisciplinary UAS research program (UFUASRG ) with the
University of Florida Department of Aerospace and Mechanical Engineering's Micro Air
Vehicles Laboratory, the UF School of Forest Resources and Conservation's Geomatics
Program, and the U.S. Army Corps of Engineers (USACE) is actively working towards
the development of a small UAS for aerial imagery collection for natural resource
assessments and monitoring applications.
The initial motivation to explore UAS applications for natural resource applications
was to save lives. Due to challenging terrain and low altitudes characteristic of aerial
surveys, light aircraft crashes are the leading cause of workplace mortality among
wildlife biologists (Watts et al., 2010). This research demonstrates the benefits of UAS
such as rapid development, amphibious operation, high spatial accuracy, high
resolution imagery and completely autonomous flight operation.
Development of the UASs was focused on for specific applications to the Army
Corps of Engineers (USACE), such as monitoring invasive aquatic plant species, and
the Florida Fish and Wildlife Service (FWS), like wildlife population research, these
among other application has been able to be performed with the developments of the
systems.
In 2011 a project (“Assessment of UAS Image and Navigation Processing
Software and Nominal Sensor Enhancements”) for the UFUASRG was funded. Where
the UF UAS would be used to take digital aerial images and produce geo-rectified
17
.mosaics, it would give data and images to the USACE to support ecosystem
restoration, invasive species control monitoring, levee safety and emergency natural
disaster response.
The Federal Aviation Administration (FAA) gave permission for certification of the
UF UAS airframe to fly in portions of the Everglades. The platform had more than 2
years of deployment and testing and the large volume of data collected on typical
missions makes automated processing attractive. The current data processing
approach requires extensive manual effort while investigations have shown that there
are ways to improve methodologies without affecting the aircraft used.
The objectives of the project include (1) assessment of appropriate software to
modernize the pre/post-processing, (2) a workflow that will include camera geometric
and radiometric calibration, (3) radiometric data processing, (4) geospatial data
processing, (5) sparse tie point generation, (6) photogrammetric adjustment, (7) tie point
densification,(8) terrain generation, (9) seam line generation, (10) radiometric
correction and (11) mosaic generation.
A second objective of this project was integration of Dual Frequency RTK GPS
receivers that permitted higher accuracy direct geo-referencing techniques which will
facilitate more efficient post-mission image processing workflows. In addition using a
high-end camera with an electronic/communication interface which allows for better
efficiency in collecting data and permits to change to new cameras without reconfiguring
the payload control software.
18
Buyuksalih and Li utilized a similar approach that the one used for the project of
this thesis, but for different and older software packages, that are also not off-the-shelf
commercial software (Buyuksalih and Li, 2003)
Unmanned Aerial Systems
UAS are remotely piloted light aircraft that can carry sensors in support of remote
sensing applications. Although the basic concept of designing small remotely piloted
aircraft has been known for decades, recent advances in miniaturization,
communications, strength of lightweight materials and power supplies have permitted
significant advances in UAS design.
The UF UAS was powered either by an electronic engine. It has a wingspan of
approximately 2 meters and flight duration of approximately 1 hour.
Navigation can be controlled by remote radio signals, usually given by an operator
who can directly observe the UAS in flight or use remote television images to view the
terrain observed by the UAS. They can also be semi-autonomous, autonomous, or have
a combination of these capabilities.
The term UAV is commonly used in computer science, robotics and artificial
intelligence, as well as photogrammetry and remote sensing communities. Other
synonyms could be, Remotely Piloted Vehicle (RPV), Remotely Operated Aircraft
(ROA), Remotely Piloted Aircraft (RPA), Unmanned Vehicle System (UVS) and
Unmanned Aerial System (UAS) can also be found frequently in publications. The FAA
has adopted the latter (UAS), which was originally introduced by the U.S. Navy.
(Eisenbeiss, 2009) Common understanding is that technology UAS represents the
entire system, including the Unmanned Aircraft (UA) and the Ground Control Station
(GCS).
19
UAS can carry cameras a variety of sensors depending on the application, and
data can be recorded for retrieval after the UAV has landed, or it can be transmitted via
telemetry to a ground receiver.
UAS remote sensing sensors include electromagnetic spectrum sensors, gamma
ray sensors, biological sensors, and chemical sensors. A UASs electromagnetic
sensors typically include the visual spectrum, the near infrared spectrum s as well as
microwave. Biological sensors are capable of detecting the airborne presence of various
microorganisms and other biological factors.
Limitations of UAVs
Current limitations include (1) initial acquisition costs for the UAS, (2) crew training
requirements, (3) limited availability of high quality and lightweight sensors, and (4) FAA
regulations for operating a UAS in the national airspace.
UAVs limit the sensor payload in weight and dimensions so that often low weight
sensors like small or medium format amateur cameras are sometimes used. UAS have
to acquire higher number of images so they can obtain the same image coverage and
comparable image resolution. These payload limitations require the use of low weight
navigation units which yield less accurate results for the orientation of the sensors. Also,
because of the nature of these artifacts, they can not achieve high flying heights.
Existing commercial software packages applied for photogrammetric data
processing are rarely set up to support UAS images therefore there are no standardized
workflows and sensor models.
Photogrammetry
Photogrammetry is the art, science and technology of obtaining reliable information about physical objects and the environment through the process of recording, measuring and interpreting photographic images and patterns
20
of electromagnetic radiant imagery and other phenomena (American Society of Photogrammetry, 1980).
Aerial Photogrammetry
Aerial or conventional photogrammetry utilizes large format imagery and ground
coordinate information to effectively recreate the geometry of a portion of the earth in a
virtual environment. In environment, reliable horizontal and vertical measurements can
be made and compiled directly into a geospatial data file. To take accurate
measurements from aerial photographic images, the following conditions have to meet:
two or more overlapping stereoscopic images cover the object to be analyzed; accurate
x, y, and z coordinates are known for at least three defined object points in the
overlapping photographs; and a calibrated mapping or metric camera is used to take the
photographs. The compilation of planimetric features and topographic information from
the photographic sources is accomplished through the use of digital stereoscopic
instruments. Digital photogrammetric workstations require specialized software and
hardware for viewing a pair of stereo images. An experienced operator can link the
images with the ground control to collect precise horizontal and vertical coordinates for
a point, line, polygon, or surface. The photogrammetric workstation recreates the
geometry of the field subject through a series of mathematical operations. These
procedures require a high level of expertise and repetition to maintain the operator’s
skill. The softcopy instrument has analytical capabilities to a precisions level of under
millimeters level. Thus, high-accuracy ground control coordinate positions are needed to
fully exploit the analytical capabilities of these instruments.
21
UAS Photogrammetry
UAS photogrammetry describes a photogrammetric measurement platform, which
operates remotely controlled, semi-autonomously, or autonomously, without a pilot
sitting in the vehicle. The platform is equipped with a photogrammetric measurement
system, including, but not limited to a small or medium size still video and/or video
camera, thermal or infrared camera systems, airborne LiDAR system, or a combination
of these. Current standard UASs allow the registration and tracking of the position and
orientation of the implemented sensors in a local or global coordinate system.
Therefore, UAS photogrammetry can be understood as a new photogrammetric
measurement tool. UAS photogrammetry opens various new applications in the close
range domain, combining aerial and close range photogrammetry, but also introduces
near real time application and low cost alternatives to the classical manned aerial
photogrammetry (Eissenbeiss, 2009).
Comparison
To compare the three forms of photogrammetry described earlier, one can start with
the planning, which in AP it usually is semi-autonomous, in UAS it is automatic and
manual; the data acquisition could be assisted or manual for AP, and autonomous,
assisted or manual for UAS. For AP the project area size is several square kilometers,
and square meters up to square kilometers in UAS. Resolution and GSD wise, UAS
having within millimeters and meters resolution and centimeters to meters for. About
distance to object and sensor to object orientation AP being 100 meters to 10 kilometers
and meters to kilometers from object in UAS Photogrammetry, both normal and oblique
case are used for the two types.
22
The AP has an accuracy of initial values of centimeters to decimeters, its normal
block size is between 10 to 1000 images, and UAS has a centimeter to 10 meter
accuracy and 100 to 1000 images per block.
Automatic Tie-point Extraction
One of the most complex and time consuming processes in photogrammetric
workflow is the extraction of corresponding points in two or several images. Tie point
extraction is the first step of aerial triangulation meant for computing the positions of the
projection center and orientation of each image. The automation of the tie points
extraction is challenging, especially when nonstandard approaches such as cameras
mounted, usually digital and non-metric, on UASs or oblique imagery are used (Shragai
et al., 2012) Another reason for the increase in interest is the larger number of images
produced for photogrammetric studies, in part by the use of the before mentioned digital
cameras and strong overlaps (notably interstrips) on the other hand, generate images in
greater number than in the past.
A tie point is a point whose ground coordinates are not known, but is visually
recognizable in the overlap area between two or more images. The tie points can also
be measured manually. Tie points should be defined in all images. They should show
good contrast in two directions, such as the corner of a building or a road intersection.
Tie points should also be distributed over the area of the block (Erdas, 2011).
Automatic tie point extraction is one of the major focal points of research and
development in photogrammetry in recent history. The automation of aerial imagery
photogrammetric processing aims at the production of tie points and their use in two
domains: the aerotriangulation itself and the automatic realization of index maps, which
is a preliminary step in any triangulation process (Shragai, 2012).
23
This automatic tie point extraction is in special interest for this research, first
because without the use of GCPs, we have to rely completely on interior and exterior
orientation parameters to measure tie points and consequently match images together
on the mosaics, and second because it is a good measure of software productivity,
according to speed and accuracy of the entire processing, where this part may dictate
the main difference between programs.
There are different aspects of the scene that can actually impact the point
extraction and these are variation of aspect of objects, where the can vary depending of
the position of the object in relation to the sensor. Forest, shades and textures, these
three together create problems in image analysis. Trees cause problems of precision of
pointing; shades move between two image acquisitions, notably between two strips,
even though they are easy to delineate they are not valid in a photogrammetric point of
view. Textures can lead to problems of identification. It is easy to confuse two very
similar details in zones with repetitive motives. Relief, this provokes notable changes of
scale and variations of discrepancy. When looking for homologous points, the space of
research is generally larger (Ghosh, 2005).
For the study of tie point extraction one have to take into account 3 different
factors: reliability, precision and minimum number of tie points. Most of the algorithm
and methods use least square adjustment technique which is sensitive to aberrant
values. Therefore one of the objectives is to find a method that provide points that are
exempt from mistakes. Second, at the end of the extraction and aerotriangulation, one
has to take into account a good model that can correct the errors in point extraction.
Finally is the number of tie points needed by the software in order to be able to later run
24
the bundle block adjustment; Erdas LPS needs a minimum of 6 points for every image,
while EnsoMOSAIC needs 6 points in each of the four quadrants in the image, and at
least 4 points per image pair.
2d3 Altimap detects the feature points, then analyzes the image set connectivity
determining feature point correspondences guided by GPS data. The image matching
strategies incorporated in Erdas LPS for tie point generation include the coarse to fine
matching, feature based matching with geometrical and topological constraints, a
simplified method from structural matching algorithm (e.g., Wang, 1998), and least
square matching for high accuracy of tie points.
Aerial Triangulation Techniques
Aerotriangulation is the term most frequently applied to the process of determining
the X, Y and Z ground coordinates of individual points based on photo coordinate
measurements. Today, the accuracy of the ground coordinates using these techniques
are within decimeters depending on the payload used (Krystek et al., 1996).
Aerotriangulation is used for many purposes in photogrammetry and for most of
the applications; the minimum number of control points is 3. For large mapping projects,
the number of control points increase. The use of Real-Time Kinematic (RTK) GPS in
the aircraft to provide coordinates of the camera at the instant each photograph is
exposed. This technique has eliminated the need for ground control (Wolf and Dewitt,
2000).
2d3 Altimap uses a two-dimensional (2D) block adjustment algorithm, then ortho-
rectifies and geo-register the mosaics applying auto level correction and color
balancing.
25
Erdas LPS uses Bundle Block Adjustment techniques that uses itself collinearity
condition as the basis for formulating the relationship between image space and ground
space. The collinearity equations are solved using least-squares adjustment to (1)
estimate the values of exterior orientation, (2) estimate coordinates of tie points, (3)
estimate interior orientation and (4) minimize and distribute data error through the
network of observations. LPS also allows the interior orientation parameters to be
analytically calibrated with its self-calibrating bundle block adjustment. LPS offers robust
error detection methods within the triangulation process to eliminate gross errors.
Bundle Block Adjustment
Almost all analytical aerotriangulation methods consist of writing condition
equations that express the unknown elements of exterior orientation of each photo in
terms of camera constants, measured photo coordinates, and ground coordinates.
These equations are solved to determine the unknown orientation parameters and
subsequently coordinates of pass points are calculated.
The bundle block adjustment allows the orientation of a block of an unlimited
number of photographs using only three GCPs. This requires that relative orientation of
the individual images within the block first established by additional tie points/image
points with unknown ground coordinates which appear on two or more images and
serve as connections between them, as shown on Figure 1-1. The tie points can be
identified manually or with automatic matching procedures. The term bundle refers to
the bundle of light rays passing from the image points through the perspective center to
the object points.
26
The exterior orientation parameters has become important in bundle adjustment
with the use of airborne GPS control and Inertial Navigation Systems (INS) which have
the capability of measuring the angular attitude of a photograph.
Use of GPS in Aerotriangulation
As mentioned before, the aim of aerotriangulation is to reduce as much as
possible the requirement of field measures, by processing simultaneously the geometry
of several images. Aerotriangulation will adjust the measures of coordinates in the
images of the homologous tie points, known ground control points, auxiliary measures
recorded during the flight, or even satellite trajectography data.
Digital photogrammetry itself brought up the measure of tie points to be automated
completely and the development of digital cameras, as they require more images to
cover a the project area, led to the search for more efficient auxiliary measures in order
to restrict again the ground measures. All these led to the implementation of highly
accurate GPS control on aircraft for the use of aerotriangulation methods.
The precision needed on the GPS observation need to be at least as good as that
of the points to localize on the ground. These precisions can vary between 5cm and
30cm, and with sometimes high speed of the planes, that means a good
synchronization between camera and GPS receiver. It is necessary that the camera
provides a signal perfectly synchronous with the opening of the shutter. Also, with GPS
measures every second one has to interpolate the position of the camera at the time of
the signal of synchronization.
The UFUASRG is working on the implementation of RTK GPS antennas on the
aircraft. These RTK modes which measure the phase on the signal carrier. The
initialization can be done with a time duration permitting the ambiguity resolution and
27
fixes it for the remainder of the mission, which the reliability is low, because due to the
aircraft turning there is interruption of the signal, making it impossible to solve the
ambiguity again on the flight. The other solution is to solve the ambiguities On-The-Fly
(OTF), which for conventional photogrammetric flights can be a limitation because of a
distance of at least 20 to 30 kilometers to the base, something that is not of a problem
to UAVs. With the use of OTF solution, it is necessary to put a GPS station close to the
zone of the photograph, which can be a implementation of a lot of time due to
equipment set up, other solution is the use of a permanent GPS station close to the
project.
Determination of the Attitude of the Aircraft
For a great measure of the attitude of the plane it is needed several GPS antennas
with perfect synchronization and separated around the plane, for better precisions. At a
higher cost but a better solution for the SUASs, due to lack of payload weight capability,
is an Inertial System or Inertial Measurement Unit (IMU) integrated with the GPS unit, it
can create a very accurate attitude measurement.
28
Figure 1-1. Principle of bundle-block adjustment. The relative orientation of the images in the block is established by both tie points and GCPs, the absolute orientation of the block within the ground coordinate system is realized using GCP coordinates. Source (Aber et al., 2010).
29
CHAPTER 2 LITERATURE REVIEW
UAS Uses and Applications
The adoption of remote sensing using UAVs in archaeology is proposed by
Eisenbeiss, (2009). The main purpose is to document archaeological sites, and to
provide better resolution imagery. The accuracy requirements are not very high,
although it has been shown that elevation accuracy using a helicopter UAV and a
consumer digital camera yields elevation models that are comparable to ground laser
scanner measurements.
Vegetation monitoring was successfully done using UASs. A Hale UAS, Pathfinder
Plus was used to demonstrate this on a coffee plantation in Hawaii (Herwitz et al.,
2004), similar to others being use to study rangelands and has been considered to be
an integral part of farm equipment (Rango, 2006).
Rapid response imaging using UASs has received a lot of attention as well. This
has been demonstrated for road accident simulation (e.g., Haarbrink et al., 2006) and
also for forest fire monitoring (e.g., Réstas, 2006).
Qingyuan et al., (2011) proposed a new UAS image mosaicking method which
uses the homogenous points extracted from the imaging stitching. The fast mosaicking
of UAS images contemplates an alternative to deal with fast mapping applications, such
as disaster monitoring, human rescue (Qingyuan et al., 2011)
The UFUASRG has used small UASs for restoration monitoring, wildlife surveys
(bird detection and counting, manatee surveys, bison survey) , habitat assessments,
forestry stand analysis, vegetation surveys (imagery can be used to measure area of
30
wetlands, vegetation coverage, habitat type, health of stands, restoration progress),
storm damage assessments (Watts,et al., 2010)
Other applications of UASs are power line inspections and surveying, where
power lines and power corridors are mapped; snow cover and snow depth; ocean color
where high resolution hyper spectral measurements enable measurements of
chlorophyll and primary production in the ocean, Digital Elevation Models (DEM),
mapping, atmospheric and meteorological measurement, environmental monitoring, like
oil spills, flooding and algae blooms (Norut, 2012)
UAV-based photogrammetry has been used for accurate 3D mapping in mine
areas. The workflow included ground control network design, image acquisition, 3D
mapping and information extraction (Liu et al., 2012).
The use of airborne differential GPS, with an accuracy of 2 to 4 cms, had been
compared to the accuracy of using GCPs in UAVS photogrammetry, with similar values
for both techniques, between 10 and 15 cm (Turner et al., 2012).
Advances in Automatic Tie Point Extraction and Aerial Triangulation
Photogrammetric aspect, signal based matching or area-based matching, is a
method that determines the correspondence between two image areas according to
similarity of grey level values, cross relation and least square techniques are known
methods for this kind of matching, the counterpart to this methods is the necessary to
use perfectly oriented images with not much rotations (Erdas, 2010).
Area-based matching is also called signal based matching, which determines the
correspondence between two image areas according to the similarity of their gray level
values. Least squares correlation techniques are a well-known methods for area-based
31
matching. Least square correlation uses the least squares estimation to derive
parameters that best fit a search window to a reference window. This technique has
been investigated thoroughly in photogrammetry (Ackermanm, 1983; Grun et al., 1988;
Helava, 1988).
Kenned and Cohen, (2003) contemplated that correlation-based matching
approach has great potential for use in general satellite remote sensing, citing that the
grid of tie-point pairs produced by the this technique is regular, it is optimal for capturing
the geometric relationships of images, moreover, the technique is repeatable, ensuring
that image libraries built up over time have consistent geometric properties. It is
relatively robust to simple distortions and inaccuracies. (Kennedy and Cohen, 2003)
Feature-based matching or signal aspect matching, it determines the
correspondence between to image features, the feature points calculated with this
methods are commonly called interest points. One operator is the Forstner Operator
(Forstner and Gulch, 1987). LPS uses this operator where the image features must
initially be extracted, and later the attributes of the features are compared between the
two images.
Another automatic tie point technique relation-based, also called structural
matching (Vosselamn and Haala, 1992), is a very which uses image features and the
relationship between the features although this approach is very time consuming.
Wang developed a structural matching algorithm, where a fully automated
matching of image features is realized without any a-priori information, even with
images from amateur digital cameras was achieved (Wang, 1998)
32
Also, a new approach using linear features has being evaluated, contemplating
that a typical aerial scene contains more linear features than well-defined points, and
that control information from an object is more reliable than individual points. (Schenk,
2006).
Tree search methods has been developed and assessed on the applications and
implications that could have in digital photogrammetry problems like object location and
recognition, stereo image matching, edge and line following and geometric reasoning
(Vosselman, 1995).
A tie point matching algorithm using least squares image matching techniques for
UAS using video imagery, therefore the resulting images are very close to each other,
meaning very high overlaps between images. Instead of searching an entire image or a
large portion of an image for a conjugate point, the search was reduced to a subset of
the image based on the point’s coordinates in the previous image (Wilkinson, 2007)
Evaluations on aerial triangulation methods had been done, including a
comparison between different additional parameter models, and an assessment of
human performance versus computational performances (Tang et al., 1997).
Self-calibrating bundle block adjustment methods use additional parameters in the
triangulation process to eliminate the systematic errors. Self-calibrating methods are
studied in (Granshaw, 1980) and (Konecny, 1994).
A procedure for automatic absolute orientation using aerial photographs and a
map, an automated system for exterior orientation which first involves automatic relative
orientation. This method has potential for use where existing maps are available and
where sufficient detail is present on the image and on the map to ensure a large number
33
of well distributed points which can be used for orientation (Morgado and Dowman,
1997).
Other approaches like the of high redundancy in the multispectral aerial sensor
input images to generate a land use classification, a DEM and true ortho-images
(Zebelin et al., 2006).
Heipke, (1997) approached the automation of interior, relative and absolute
orientation with a more primitive way, using scanning of non-digital images, stating that
is a more reliable technique than GPS and INS technology of his time.
34
CHAPTER 3 METHODS
Study Area
A data set was collected utilizing the (UF UAS) payload integrated onboard the
UAS in order to evaluate minimum overlap and side lap required to create mosaics with
the three photogrammetry software programs assessed (2d3 Altimap, EnsoMOSAIC,
LPS). A flight plan consisting of ten parallel flight lines equally spaced twenty meters
apart was prepared at Archer Fields, Archer, Florida. Imagery was collected at a rate of
one exposure every 2.5 seconds. Each flight line was approximately 800 meters in
length and orientated in a north-south direction. The flight plan was executed in a back
and forth pattern; flying south to north on line one and returning from north to south on
line two. This pattern was repeated for the ten lines. Upon completing flight line ten, the
flight plan was flown again; back and forth, in the opposing direction (flying south to
north on line ten and returning from north to south on line nine). This pattern was
repeated for the ten lines back again in order to minimize the effects of inconsistent
overlap due to wind speed. This flight plan was able to create different scenarios of
overlap and sidelap coverage, for assessment of the software in different conditions. In
Figure 3-1, shows where the number of strips are shown with wind direction and
geotags of images on capture.
The following procedures were utilized to vary the exposure interval for this
investigation: (1) process all data along the planned flight path (all flight lines, both
directions); (2) process only exposures collected in the upwind direction along the
planned flight path (all flight lines, north to south direction only); (3) process only
exposures collected in the downwind direction along the planned flight path (all flight
35
lines, south to north direction only); (4) process only exposures collected in the upwind
direction along the first back and forth pass of the planned flight path, utilizing every
other line (even numbered lines, north to south direction only); (5) process only
exposures collected in the downwind direction along the second back and forth pass of
the planned flight path, utilizing every other line (even numbered lines, south to north
direction only); (6) process only exposures collected in the first back and forth pass of
the planned flight path (all flight lines); (7) process only exposures collected in the
second back and forth pass of the planned flight path (all flight lines); (8) process only
exposures collected in the upwind direction along the second back and forth pass of the
planned flight path, utilizing every other line (odd numbered lines, north to south
direction only); (9) process only exposures collected in the downwind direction along the
first back and forth pass of the planned flight path, utilizing every other line (odd
numbered lines, south to north direction only). For every step of this process explained
above, a block was created in order to run separately all datasets on the software to be
assessed. Table 3-1 shows the specifications of separation and overlap for every block,
and Table 3-2 shows the lines and directions used for every block. On Figure 3-5 is a
map of all 9 block configurations.
Determination of Ground Elevation on Center of Images
In order to get the elevation on the ground, on the exact position of the center of
the images on the moment of capture, a five-foot contour line layer was downloaded
from the Florida Geographic Digital Library (FGDL). This layer was clipped to the
boundary of City of Archer. The contour line shapefile was later converted into a raster
surface within the study area using the cleaned flight lines and geotags of the images
taken with the UAS, which in result is a DEM of the study area shown in See Figure 3-2.
36
Later, the DEM was interpolated to the spatial position in the projection of the geotags,
resulting in a new shapefile were the previous geotags file were added a new feature
called elevation.
Calculation of Overlap Percentages
An important element in photogrammetry, is overlap percentages, both in forward
and side, this is to permit any links and tie between adjacent images. Assessment of the
impact of different overlap percentages on mosaic processing was made. The forward
separation between flights was inconsistent due to wind gusts along the flight line. This
was one of the reasons to use this flight direction to only have the incidence of wind on
one direction.
The dataset was in WGS 1984 coordinates. To calculate distance in meters
between images, the coordinates where transformed to UTM Zone 17 North projection,
and distance was calculated between consecutive images along the flight path using a
distance matrix. This distances where averaged per block configuration. The ground
distance of an image side can be calculated with Equation 3-1, where S=ground
distance of image S´=sensor size in millimeters, Hg= flying height, f=focal length.
(3-1)
Equation 3-1 was used to calculate the ground side of every image. And then
calculating an approximate overlap percentages using Equation 3-2, where S1=ground
distance of Image, and S2 = ground distance of adjacent image. These values were
averaged for all the blocks.
(3-2)
37
Tree and Cloud Coverage per Image
In order to estimate the impact of tree and cloud in photogrammetry processing, a
manual survey was done to the entire dataset. A percentage of tree and cloud coverage
was measured for every single image.
Workstation Specifications
The workstation used for this evaluation was an HP z800 Workstation, with an
Intel Xeon W55880 @3.20 GHz 3.20 GHz Processor and a 48.0 GB Memory Ram, in a
Windows Vista Professional 64-bit Operating System.
UF UAS Payload Output Data
Each flight automatically generates a folder labeled with the time and date. The
folder contains two elements, the image files and a log file. The format of the image files
is by default .jpg, although the image format is selectable. The log file is generated in
real time and combines the information from all payload sensors but the images into a
single ASCII file. The log is formatted with each line corresponding to a data or status
packet, prefixed by a 3 letter code indicating the source of the packet. The log file is
parsed and the packets processed to produce an output file, the geotags file, which
provides the direct georeferencing parameters associated with each image. The
parameters are calculated by interpolation of the navigation packets using a “Burredo”
synchronization packets associated with the image exposure. The “Burredo” is
synchronization device, manufactured by the UF team, designed for synchronization
between a the Olympus E-420 and an INS/GPS. The “Burredo” allows for
synchronization of a wide range of sensors with minor modifications to the signal
conditioning circuitry to handle the voltage level. (Perry, 2009).
38
The geotag file is the one that was used for all the post-processing, it contains:
Image name and folder path; longitude and latitude, in decimal degrees; ellipsoid height,
in meters; and the image orientation (pitch, roll and yaw), in the form of omega, phi,
kappa in degrees.
For the omega, phi and kappa values, we have to visualize XYZ coordinate
system in the origin of the focal point (center of image), omega is a rotation about the
photographic x-axis, phi is a rotation about the photographic y-axis and kappa is a
rotation about the photographic z-axis as shown in Figure 3-3.
Non-Vegetated Area of Study
The forested area in Nortrth side of Archer Fields Area has dense vegetation,
making impossible to process data on that zone with Erdas LPS, therefore a smaller
non-vegetated area of study (270 meters long, 120 meters wide) was created for every
block keeping the forward and side overlap percentage values, to be able to make a
comparison between LPS and EnsoMOSAIC. The Area has (270m x 120m). The
clipped area is shown in Figure 3-4.
39
Table 3-1. Block Configuration and Properties
Block
Distance (m) Overlap (%) Average (m)
Configuration Images Fwd Std Side Fwd Side Ground Height X Grnd Y Grnd
0 335 24.98 12.22 20 73.86% 72.07% 25.05 162.88 95.47 71.60
1 221 33.98 4.76 20 62.26% 72.03% 25.05 162.68 95.33 71.50
2 114 72.75 5.87 20 27.86% 72.14% 25.04 163.25 95.73 71.80
3 113 35.13 4.43 40 63.11% 43.92% 25.10 162.39 95.10 71.32
4 57 68.31 5.93 40 29.13% 44.63% 25.02 164.08 96.32 72.24
5 170 48.26 17.17 20 49.34% 71.96% 25.09 162.40 95.11 71.34
6 165 47.41 15.77 20 50.62% 72.17% 25.01 163.37 95.83 71.88
7 108 36.88 4.96 40 61.45% 44.20% 25.01 162.99 95.58 71.68
8 57 69.91 5.79 40 26.53% 43.95% 25.06 162.42 95.15 71.36
Table 3-2. Block Configuration Flight Lines and Directions Block
Configuration Direction Flight Path Flight Lines Downwind (North) Flight Lines Upwind (South)
0 All Both 1, 2, 3, 4, ,5, 6, 7, 8, 9, and 10 1, 2, 3, 4, ,5, 6, 7, 8, 9, and 10
1 Upwind Both None 1, 2, 3, 4, ,5, 6, 7, 8, 9, and 10
2 Downwind Both None 1, 2, 3, 4, ,5, 6, 7, 8, 9, and 10
3 Upwind First None 2, 4, 6, 8 and 10
4 Downwind Second 2, 4, 6, 8 and 10 None
5 All First 1, 3, 5, 7 and 9 2, 4, 6, 8 and 10
6 All Second 2, 4, 6, 8 and 10 1, 3, 5, 7 and 9
7 Upwind Second None 1, 3, 5, 7 and 9
8 Downwind First 1, 3, 5, 7 and 9 None
40
Figure 3-1. Map of flight mission. Including of the satellite imagery of AOS, wind
direction, Strip numbers.
Figure 3-2. Contour Line Data, Flight Plan and Digital Elevation Model. A view of the Contour Line Shapefile downloaded from FGDL along the image captures and the Digital Elevation Model created using the ArcGIS 10 software.
41
Figure 3-3. Image Exterior Orientation Parameters: Omega, Phi and Kappa.
Z
X
Y
Flight Path
42
Figure 3-4. Non-vegetated area of study.
43
Figure 3-5. Block Configurations.
44
CHAPTER 4 RESULTS AND DISCUSSION
2d3 Altimap
2d3 Altimap is an inexpensive solution to solve mosaicking for low altitude flying
aircrafts. This solution has the ability to correlate the images in “image space” by the
images themselves alone. If there is EXIF header about the GPS position of the image,
the software will geo-locate the resulting mosaic.
2d3 Altimap automatically compares features in each image with other images to
determine their match in image space, a process called Wide Baseline Matching,
making 2d3 Altimap great for processing even if no data is available from the imagery. If
the image dataset provided by the user does not have enough overlap, 2d3 Altimap will
create separate mosaics for cases where there is no overlap in the image sets.
Altimap will read JPG, BMP, PNG and GIF imagery files provided by the user, and
NMEA, XML, EXIF and TXT data files. Altimap will output mosaics in JPG and PNG
imagery files, and also KML, Tile Map Specification and XML creating an output mosaic
that can be exported to Google Earth, KML File, and it can be viewed on the current
location of the imagery, just like geolocated orthoimages.
The different capabilities of the Altimap software are mosaicking, which by using
highly accurate image registration it will stitch together adjacent images, and use them
to create a large mosaic. This capability can also be used with video recording in an
aircraft; two-dimensional feature extraction, which is that the program will detect and
identify hundreds and thousands of features in a scene and follow their motion
throughout a sequence using a form of corner detection and also has the capacity to
match and track shapes as they change in two dimensions with their relative orientation
45
to a moving camera. This can be helpful in the identification of fiducial style markers.
The camera’s tracking technology makes it possible to calculate the path of the
originating camera in three dimensional space and describe the three dimensional
position of two dimensional features within the source image sequence. For a stream of
multiple still images, the structure from motion approach involves the automatic
identification of hundreds and thousands of distinctive points that appear in areas of
high contrast or high texture. Structure from motion enables the three dimensional
movement of the camera to be inferred from the 2d motion observed in the image
sequence. Visible in the image above is a red line indicating the inferred trajectory of the
camera in 3d space and the camera view frusta for some of the frames.
2d3 Sensing
2d3 Sensing is a remote sensing company that specializes in computer vision
softwares and solutions for imagery, metadata acquisition and processing. 2d3 started
to adapt their technology for use in aerial imaging applications, developing products
covering a wide range of real time and off line computer vision capabilities for
processing of aerial motion imagery. Designed from the ground up for analyzing and
processing motion imagery, these products won the company a lot of contracts with the
military and commercial business for remote sensing and security.
2d3 acquired Sensing Systems which had developed and fielded a motion imagery
software development media toolkit. 2d3 Sensing, the resulting combined entity, offers
a complete spectrum of software technology for the management, enhancement,
exploitation, and dissemination of imagery and metadata.
46
2d3 Data Preparation
The 2d3 Altimap works taking images with EXIF data written to them, with
longitude, latitude and height information. The image that Altimap uses has to be at a
specified directory, and the output data will also be located in this folder. The Altimap
software is the easier of the three programs assessed to work with. The one with less
steps is written in a CMD format script, and to facilitate the handling, a python script was
written to do some of the preliminary data preparation on one script.
Image Cleaning
In order to use the correct image data to be processed with 2d3 Altimap, the first
step is to clean and create blocks using the geotags file to create feature points in a GIS
software like ArcGIS or QGIS. To start the filtering process, we can start after the first
hundred or so images usually, which are the images taken during takeoff and while the
plane gains enough height to start its flight over the study area according to the flight
planning already created, and the last pictures which are clearly during plane landing.
The next step in the cleaning process is to determine what was the study area
already planned before, and delete all the pictures taken out of the study area while the
plane is curving to return to the planed flight lines. A good strategy is to identify where
the straight lines are starting to break, and where the orientation angles are getting
greater than during the normal flight line. After all of the correct images are selected,
copy the image file paths from the feature table, and create a CSV file with it, to be used
later, called BlockXX.csv(XX for the number of the block created).
Block Creation
The 2d3 Altimap is designed to process up to 250 to 300 images taken from
SUASs as stated by its developers. To evaluate this capability, an experiment using
47
Sea Horse Island in Cedar Key was done. The mission was on the North side of the
island. It’s an area approximately 1,500 meters by 360 meters or a total of 330 images
after cleanup.
The flight plan was executed in a back and forth pattern; flying east to west on strip
one (farther south) and returning from west to east on strip two. This pattern was
repeated for five lines, where the pass returned on the same line. For this experiment
only the downwind flight path was used on strip five, considering for worst case scenario
data with less overlap percentage. Then the pattern continues from strips six with east
to west bound until strip thirteen with a west to east bound. The first 9 strips, which are
around 260 meters in the north direction, are around 1,500 meters long. The last 5,
around 100 meters wide, are 1,100 meters long. This is done like these because of the
configuration of the island of the study area. For this dataset the average side overlap is
65% and forward overlap ranges from 40% in downwind strips to 80% in upwind strips.
For this experiment, 4 different block configurations were created to get different
scenarios (number of strips, number of total images, overlap, features on final mosaic).
Configuration 0, which is the entire image set of the study area; configuration 1, the
division of the 13 strips into 3 blocks of 5 strips each, keeping one common strip
between adjacent blocks to leave certain overlap between blocks; configuration 2, using
two rectangular blocks, depending on the length of the different strips, leaving the first 9
strips which are 1,500 meters long, and a second block with the last 5 strips adding strip
number 9, with their approximate length of 1,100 meters long; and configuration 3,
which the block was created from northeast to northwest then moving southeast to
southwest. This was to create more potential similar geometries and image quantity
48
between blocks, using primarily 5 strips on the vertical direction, and approximately 9
images (upwind direction) on the horizontal direction, the final block was approximately
380 meters by 130 meters, 3 on the top tier, 4 on the middle tier and 4 on the bottom
tier, 11 blocks in total. Block Configurations of Sea Horse Key Mission are shown in
Table 4-1 and Figure 4-1.
Total Processing Times
After adding up the total time of processing for every configuration (See Table
below), the total time of processing is similar for every configuration. This favors
creating just one block of data for the study area with datasets similar to the one used
for this experiment with specifications described above. Due to the data preparation
time for every block, it greatly reduces total processing time creating the one block
instead of a block per certain amount of square meters that would be the other option.
Table 4-2 shows total processing times for all configurations.
Processing Time Results
The processing time for every block ranged from a few minutes to close to 20
minutes (As shown in Table 4-3), which gives this software a reasonable time for
processing mosaics. With this amount of time, the UAS can be deployed and extract the
data from it, and the entire mosaic processing can be done in around one hour including
the data preparation explained above. Which makes it very convenient for almost real-
time monitoring of cases that need a really fast aerial image assessment (like disaster
assessment or emergency monitoring).
Block32 and Block35 weren’t able to be processed to a mosaic due to the
complexity of the features in them such as water with waves. The program ran for less
than one minute in each case, helping us by knowing that when a mosaic does not have
49
enough image pairs to create a mosaic, it will not try to match the image indefinitely.
Block0 is also on the table, which is the Block that had the entire Strip 9 flown twice
both downhill and uphill, so we can evaluate this block in processing time for total image
count before processing.
By comparing total images contained in each block versus processing time (See
Figure 4-3), we can see there is no a true correlation between images and processing
time. There is certain relationship, accounting for a big difference in image count, but
not a true one. The user can not say from the beginning, I have X amount of images so
this will take X amount of time to process, and could be able to decide on going for
bigger or smaller blocks accounting with that specific number. A value of total
processing time per image was calculated on Table 4-3, showing again the lack of
relation between these factors.
This lack of correlation explained above is notable even in blocks with the same
number of strips, or the same amount of overlap percentage, or even same amount of
images. On Figure 4-4, it can be seen that from Figure 4-3, but zoomed in to Block
Configuration 3, where it shows an entire non correlated result all over the chart. This
result is perhaps because, when taking smaller blocks, the blocks had only one feature
in them (water only, bare ground only, vegetation only or grass only). Unlike bigger
blocks that probably contained all features (water, bare ground, grass and vegetation) in
them, making them easier or harder to be matched by the software. Figure 4-5 is
another example of the lack of correlation between overlap percentage and processing
time. This time the chart is divided in the different Block Configurations, because here
50
only blocks with similar numbers of images can be compared to others regarding
overlap and time.
On Figure 4-6 and Figure 4-7 there is enough evidence that the final output size or
quantity of images matched in the final mosaic has a great relation with processing time.
Unfortunately, this is not good for stating a good idea from before the process on the
quantity and size of blocks to be created for the processing.
Maximum Quantity of Images Processed by 2d3
To establish the maximum number of images that can be processed, there was
another mission adjacent to this 13 strips of data, to the south (south side of the island),
and one strip was added to the 13 original, with approximate of 1,500 meters long each,
for every run, the software worked well until the number reach 600 images, after that the
software started to create wrong shaped mosaics, as shown on Figure 4-2, and after
650 images it started crashing due to high number of calculations.
Import Exif Data to Images
Exif, Exchangeable image file format, is a standard that specifies the formats for
images, sound and ancillary tags used by digital cameras, scanners and other systems
handling image and sound files recorded by digital cameras. Almost all new digital
cameras use the Exif annotation, storing information on the image such as shutter
speed, exposure compensation, F number, what metering system was used, if a flash
was used, ISO number, date and time the image was taken.
The Exif format has standard tags for location information. In these days a growing
numbers of cameras and mobile phones have a built in GPS receiver that stores the
location information in the Exif header when the picture was taken. Other cameras that
don’t have a built in GPS receiver are not compatible with a separate GPS receiver that
51
can be adapted to the flash connector and can write the same geolocation information
to its images. For other cases users can add to any digital photograph manually or from
a handheld GPS receiver to the images taken with digital cameras, this process is
known as geocoding.
To start the CMD script to prepare to work with 2d3 Altimap, 2d3feeder.py, the first
step is to write the Exif data to the images. In order for this to work, we have to make
sure on the geotags.csv file, parsed from the log file, that there is no negative elevation.
This geotags file already has the original directory where all the images are located. So
the CMD script will automatically write the Exif info on the original images; longitude,
latitude, height, and orientation data. Figure A-1 and Figure A-2 show webpages that
can be used to read the full Exif data, and also shows on Google Maps the location of
the image, like exifdata.com or http://regex.info/exif.cgi/exif.cgi.
Creating Image Directory
On this step of the data preparation, the hard work is already done on the data
cleaning process. The next step is to subset al.l images to be used to a new directory,
which is done with the geotags.csv file created on the cleaning process. This file
contains nothing more than the image file path. It will create a new directory at the
location we select, to be called Blockxx (xx is the number which the block was called).
Processing
After the images with the Exif info have been written to the new directory,
everything is ready for post processing. Now a scale and a desired output format need
to be chosen.
Changing the settings can be done in the 2d3feedyr.py script, on the last line. First,
change the scale factor from 0.1 to 1.0, the default of which is 0.25. To change the
52
output settings, just write the command that fits more with the desired output: --single,
which outputs a single Geotiff image plus a KML wrapper; --tiled_kml, the output is a
Geotiff tile set. Other processing options are an exhaustive image matching method and
a bundle adjustment method.
Processing Report
The 2d3 software was the most difficult to use of the three software programs
tested. It does not contain a user’s manual or help menu. Also it is in a command line
interface.
The software reports its processing steps and results, but this report comes in a
DebugDump approach which can be found in the directory where the Altimap.exe file is
located. This debug dump actually is every line written by the software during
processing. It contains all of the coordinates read from images, the feature matching
process, and rectifications.
Processing Time Assessment
The intention for this work was originally to do a time processing assessment and
comparison between all software evaluated, but the LPS and EM software require
manual user input in different steps of processing so it was difficult to quantify time in
these programs. In 2d3 Altimap the software runs all the steps altogether and with no
manual user input after the processing starts. A stopwatch was used to calculate the
processing time for every block configuration, and to determine what type of relation
there is between processing time and different specifications of the datasets.
The processing time has a direct relationship with the quantity of images on the
block, as shown on Figure 4-8. This means that for constant features on the images, the
total processing time can be predicted by considering only the image count.
53
Overlap Percentage Assessment
The intention of running the flight lines several times and in different directions was
to get different scenarios of forward and side overlap percentages, and to evaluate the
impact of these on UAS photogrammetry processing.
The images processed per minute ratio has no relation with different overlap
percentages, as shown on Figure 4-9. In this figure, the side lap percentages were
divided into two groups, 40% and 70%. With the side lap percentages fixed, the forward
overlap impact of image per minute of processing is not related.
As mentioned before the photogrammetry mosaic processing is a multi-step
process. 2d3 Altimap matches tie points within adjacent images, and even though 2d3
does not detail every single tie point, it does give results on pairs of images matched.
To visualize the difference that forward overlap percentages have on image pair
matching, a scatter plot is shown on Figure 4-10, where a clear increase of forward
overlap means an increase in pairs correctly matched. We can see that with a greater
side lap percentage, there is a greater percentage of image pairs matched, showing a
strong impact of overlaps in the image matching technique used by 2d3 Altimap.
Finally the scatter plot on Figure 4-11 shows an increase in forward overlap means
an increased percentage in images mosaicked.
Cloud and Tree Impact Assessment
Using the manual survey of all the images where percentages on tree and cloud
coverage was created. To establish what the presence of trees or clouds in the datasets
can do to change the mosaic processing, a percent of images with the presence of trees
and the presence of clouds, and these were compared on each block to the percentage
54
of images mosaicked to see if an increase of images with trees or clouds means a
decrease of images mosaicked percentage.
After an in-depth evaluation on Figure 4-12 the scatter plot does not show he
expected results that the higher percent of images with trees or cloud will decrease the
percent of images mosaicked.
EnsoMOSAIC
EnsoMOSAIC is an aerial digital imaging and image processing system which
produces, geo-referenced multispectral image mosaics that cover large land areas. This
software also has the hardware counterpart for image capturing and processing for
users that need a system to start imaging and ortho-production.
EnsoMOSAIC UAV software reads aerial images taken with compact cameras
carried onboard UAS, and processes them into seamless image mosaics.
The input requirements for the EnsoMOSAIC software are; the images, in any
common format; GPS coordinates from the flight log in any autopilot model; and camera
calibration and parameters which can be calculated in another Mosaic Mill software
package called RapidCal. UAS2EM converts the flight logs from many of the
commercial available UASs and on the market.
Also, as optional parameters, to the mentioned above are, the initial image
orientation, which are the roll, pitch and yaw calculations for every image in order to
increase processing speed and improve accuracy; and ground control points (GCP),
which contribute to improve accuracy.
EnsoMOSAIC UAS processes automatically the images into ortho-rectified image
mosaics, which have map coordinates and are thus GIS-ready and viewable
immediately in Google Earth. EnsoMOSAIC applies photogrammetric principles, in
55
contrary to image stitching which is commonly used for UAS photogrammetry programs
images. Stitched images are often visually pleasing, but contain distortions which are
not suitable for area or distance measurements. EnsoMOSAIC does not stich, but rather
it rectifies images into ortho-mosaics which are free of distortion in areas with elevation
variations. This is possible because EnsoMOSAIC calculates a DEM within the
rectification process. Ortho-rectification creates ortho-photos and ortho-mosaics, which
can be used in the most demanding topographic mapping projects.
Mosaic Mill
MosaicMill is Finland based technology company established in 2009. The
development of EnsoMOSAIC tools was started in co-operation with Technical
Research Center of Finland and is today being continued by MosaicMill with its partner
companies.
EnsoMOSAIC UAV and EnsoMOSAIC 3D are software solution for UAV image
rectification and DTM and DSM processing. EnsoMOSAIC is also a turnkey imaging
solution for manned aircrafts containing a complete set of software, hardware and
support components.
Input Data
EnsoMOSAIC requires that digital images have GPS coordinates recorded in the
aircraft or estimated with maps. Camera (or aircraft) exterior orientation parameters are
not required but if they exist the rectification process is faster. Most common image
formats are accepted.
For the data input and preparation, we should define a few concepts. Area is a
geographically limited area of operation, they are mostly continuous and contain a
single block of images. Areas can also consist of more than one block of images. A
56
block of images within the area, there can be one or several block in one area. A block
can be divided into sub-blocks, this is recommended especially if the main block is
large, when computer capacity may restrict image processing. The block adjustment
may be slowed down, if the number of images in one bloc is too large for the capacity of
the processing computer.
Current day computers rarely require division of blocks into sub-blocks for
processing. It is recommended to process any block as a single calculation unit. The
processing productivity depends on the computer capacity, number and type of images
in a block, spectral and pixel selection settings and method used in the mosaic
formation. The maximum number of images that can be processed simultaneously
varies, but with modern computers of 2-4 GB of RAM image blocks of tens of thousands
frames are manageable. The limits can be found by empirically testing in each computer
setup. Blocks larger than the computer limit should be split into sub-blocks and
processed separately. A sub-block should have a regular form as possible, since
irregular shapes may lead to unwanted rectification results. Overlapping surface must
exist between adjacent sub-blocks in order to facilitate their re-joining.
Creation of Blocks
In order to use the correct image data to be processed in EnsoMOSAIC, the
dataset has to go through the similar data and image cleaning process as 2d3 Altimap
and Erdas LPS. The images captured at the ascending and descending times should be
erased, and the images captured on the turn back process while taking the next strip on
the flight.
The first step in the EnsoMOSAIC processing block is the stage in the mosaic
formation for a new area; three essential files have to be created in order to use it in the
57
EM software, Block Project File (.trp), GPS File (.gps) and a Camera Calibration File
(.cam).
TRP File
The first of the files to be created is the TRP File which is a block definition and
master file, it defines the output and location of camera and control point files, and
image file names by flight line. The TRP (.trp) file contains 8 header lines where
comments, output projection, location and name of camera calibration file, location and
name of the ground control point file, approximate mean ground elevation, rotation of
the camera, background NCA and NCM map path and file names, location and name of
background map files. After the first 8 lines, lines containing each an image and the
following information: flight line number, frame number, image file and image directory.
If the images are captured in raw formats in must be specified on this file. Camera
rotation is in degrees counter-clockwise, in relation that the aircraft’s nose is pointing.
Zero (0) for no rotation, landscape mode; 90 or 270 is for rotated portrait mode; and 180
is for backwards landscape mode. There must be a data row in the GPS (.gps) file for
each image listed in the TRP (.trp) file, the GPS file may contain more data rows than
there are images in the TRP file.
GPS File
The GPS (.gps) file is an image GPS-coordinates and optional camera orientations
file. This file contains variable columns, but the predetermined ones are as follow: Line
number, frame number, longitude, latitude, altitude, heading, UTC date, UTC time, UTC
year, yaw, pitch, roll and tilt tag.
Line and frame numbers must match those in the TRP file, longitude and latitude
must be in decimal degrees southern and western hemisphere with a minus (-) sign;
58
altitude in meters; heading in decimal degrees taking N as zero (0) degrees clock wise
to 360. The UTC date, time and year are optional and used in sun angle correction and
to calculate automatic links, if omitted sun corrections and links cannot be calculated but
all other functions are available. Tilt angles and tilt tag are also optional and if used the
UTC date, time and year are also required. The EM convention for tilt angles are: yaw,
clockwise positive degrees from 0 to 360; pitch, nose up positive degrees, nose down
negative degrees; and roll, left wing up positive degrees, left wing down negative
degrees.
CAL File
The CAL (.cal) file is a camera internal orientation and calibration parameters. This
camera calibration file contains channel, focal length, principal point’s coordinates,
general scaling and calibration parameters obtained through a proper calibration
procedure.
Pyramid Images
The first step in processing is creating the image pyramid layers, these enable fast
image display and faster calculations, these pyramids are a principal component of the
automatic aerial triangulation algorithm used by EnsoMOSAIC, for this step all pyramids
were created including level 0 for original images.
Automatic v7 Aerial Triangulation
Automatic Aerial Triangulation is an algorithm used by EnsoMOSAIC that contains
both tie point location and block adjustment altogether, it uses a measurement three-
stage process: (1) initial, (2) intermediate and (3) final. The initial assumes that the
exposure locations have some accuracy,that means that GPS observations were
collected in flight. For the initial stage the software recommends to start at level 5, or
59
level 4 for lower format images, by trying several times to start in level 4 was better
because enough tie points were found while in level 5 they weren’t. The intermediate
stage assumes that image positions and orientations are available, the first estimates of
the orientations where calculated by initial stage. The intermediate stage is run from the
next stage after initial stage down to level 0, as layer 4 was used for initial stage the
intermediate stage was run individually from 3 to 0. The final stage was used to finish
the calculations at the lowest level and to transfer the results from the lowest level to the
main level. The variation of Root Mean Square Error (RMSE) per measurement stage is
shown in Image 4-13.
The tie point assessment for EnsoMOSAIC was run for every block configuration
using 150 pixel search area and 0.80 image correlation, the criteria to accept and reject
tie points. These values were set similar to Erdas LPS, so a comparison between both
can be done; the correlation limit of 0.80 was used because the EnsoMOSAIC needs
more tie points per image than LPS (6 for every quadrant versus 6 for entire images).
Figure 4-14 shows that for higher overlap percentages the tendency was that more
tie points were generated, a result expected, also more tie points were generated while
the total images on the block increased as shown on Figure 4-15. The number of tie
points per image was less than 10 for low overlap blocks and more than 30 for higher
overlap percentages blocks.
The tie point generation accuracy was assessed using 10 manually selected
random points and later evaluated to see if they were matching on the different images
that were present. This result are shown in Figure 4-16, were not as expected forward
or side overlap percentages had any relation with accuracy, the accuracies ranged from
60
10% to 90% of tie points right. Figure 4-17 shows the percent of the images that didn´t
have enough tie points, have clouds present on them. Also the percentages of the
images that had pitch or roll values higher than 5 degrees are shown on Figure 4-17.
According to these values, the images not used had considerable more clouds than
high pitch and roll values in them.
If after the first tie point search the blocks did not converged, different search
criteria was used, including changing tie point options and the spatial location of tie
point search, when the triangulation converged, a review of residuals was done and
higher residual values were deleted, as were higher altitudes and lower latitudes. A
block adjustment had to be done after deletion of points, then more deletion of higher
residuals, higher altitudes and lower altitudes until residuals less than 10 pixels were
reached and altitude levels as expected, (20 to 30 meters for this dataset), when this
was done, another block adjustment was done and the next measurement stage was
later run, as described earlier, this is done until the final stage was complete.
Figure 4-18 shows that overlap percentages had no impact on Root Mean Square
Error (RMSE) after Aerial Triangulation converged. Figure 4-19 show that in most cases
for higher quantity of images used, the higher RMSE values were, meaning that when
less images were used for mosaicking the more accurate the Aerial Triangulation
process was.
Ortho-rectification
After the triangulation converged, the ortho-rectification process involved 2 steps:
(1) DEM derivation and (2) ortho-mosaic formation. The DEMs created had pixels of 5
meters on the side. The ortho-mosaic was created out of original images, the preferred
option for imagery collected with digital camera. DEM created previously was used for
61
the mosaic resampling, as it practically always improves the mosaic quality. The mosaic
was created using the entire area, and a spatial resolution of the mosaic introduced was
0.05 meters as estimated for the imagery collected with the Canon camera for the used
flying height.
Erdas LPS
LPS is a software application for performing photogrammetric operations on
imagery and extracting information from imagery. LPS is important because it is a large
commercial photogrammetry application that is used by numerous national mapping
agencies, regional mapping authorities, various DOTs, as well as commercial mapping
firms.
LPS is an integrated collection of tools that enable the user to transform raw
imagery into data layers that are to be used in digital mapping, GIS analysis and 3D
visualization needs. LPS provides support for air, space and terrestrial sensors. LPS
handles triangulation and ortho-mosaic production, broad area mapping, transportation
planning, engineering and facilities mapping, defense applications and close range
applications.
LPS photogrammetric and image processing algorithms for automatic point
measurement, triangulation, automatic terrain extraction and sub-pixel point positioning
help improve accuracy while simultaneously increasing productivity.
LPS supports a wide array of workflows, processing imagery from a variety of
sensors. With highly flexible, easy to use tools, LPS supports the creation of
photogrammetric data products for a variety of purposes fluctuating from GIS to
precision engineering applications (Erdas, 2011)
62
Erdas ImageStation
ImageStation is a software that enables digital photogrammetry workflows,
including projection creation, orientation and triangulation, 3D feature collection and
editing, DTM collection and editing, and orthophoto production using aerial and satellite
imagery. It differs from LPS that is designed for high-volume commercial
photogrammetry and production mapping users who need to move large quantities of
raw information to an actionable or exploitable format. This software may fulfill the
characteristics of UAS, that because lower flight altitudes, more images are acquired for
specific areas than conventional airplanes.
Erdas
In 2010, ERDAS’ parent company, Hexagon, acquired Intergraph. Both Intergraph
and ERDAS have rich histories as pioneers in the geospatial software business, with
Intergraph established in 1969 and ERDAS in 1978. ERDAS led the way with image
processing and raster handling, the ability to maximize the pixel. Intergraph built a
vector based strategy to build databases of geospatial intelligence. ERDAS and
Intergraph are now a unified business as a result of the Intergraph acquisition (Erdas,
2011).
Image Cleaning and Input Files
In order to use the correct image data to be processed in LPS, the dataset has to go
through the similar data and image cleaning process as 2d3 Altimap. The images
captured at the ascending and descending times should be erased, and the images
captured on the turn back process while taking the next strip on the flight.
The next step in the data preparation is the software input file, which the NOVA
payload is design to output the data how the LPS reads it in a CSV .csv format. The
63
way the data should be organized by columns is: ID, which is image number; Path,
where the image is located in the computer; longitude, in degrees; latitude, in degrees;
altitude, in meters; omega, in degrees; phi, in degrees; kappa, in degrees. The rotation
angles convention should be as in Figure 3-3. This file should be converted to DAT
(.dat) Format, supported by the software.
To check if the orientation angle convention are right, a good practice is to turn on 2
images and add manual tie points, and find ATPs, if the found tie points are not correct
that means that the orientation parameters are wrong, and omega, phi and kappa
values should be checked, and rotate.
After the images were cleaned and the input file is created, the next steps were to (1)
create a block in the software, (2) check the coordinate systems (3) add the sensor
used (an Olympus E-420 DSLR), (4) add the camera calibration parameters. For this
research, the camera had a focal length of 25 mm and focal point of coordinates on the
center (0,0) of the image.
In the interior orientation option, open the Dat (.dat) File input file created. The size of
every pixel should be changed in the interior orientation parameters in the Cell Array
workspace to 4.74 micrometers on each side. The images should be represented on
real coordinates on the map if, that is the case the Interior Orientation is correct.
Perform Automatic Tie Point Generation
After confirming the images on right place in the map space, the next step is to
conduct an ATP generation. In the ATP search, the tie point could be by Exterior
Orientation and or Ground Control Points, or by Tie Points already generated; search
area, initial accuracy, and matching coefficient and point density; also the pattern can
be selected for the creation of the automatically generated tie points.
64
Initial accuracy displays the relative accuracy of the initial values used by the ATP
process, although the default value is 10%, the developers recommend a 25% if the
exterior orientation option was used for initial type. Feature point density displays the
feature point density percentage based on internal default, this value was tried for 10%,
50% and 100% in the Tie Point Generation Experiment to be explained next.
Other options are the existing point transfer, which can be changed from “No
Transfer, New Points Only”, “Transfer Only, No New Points” and “Transfer With New
Points”, I will play with this options when I need them, to re-do a ATP generation when
not enough points were found for some images. Other options that were kept in their
default value for this work Correlation Size (7 x 7), Default Distribution (5 x 5), Defined
Pattern (Default Off), Intended Number of Points / Image (Pattern), Keep All Points
(Off), Starting Column, Column Increment, Starting Line, Line Increment.
Tie Point Generation Experiment
To establish the better options a small experiment was conducted, using the block
configuration 1, the ATP function was run several times using different search area
criteria (See Table 4-4), point density percentage and initial accuracy values. An
analysis was conducted evaluating the quantity and accuracy of tie points with every
configuration, in both grass area and the forest area. 10 manually selected random
points were taken from the grass area and 10 manually selected random points from the
forest area, all 20 points were evaluated to see if they were matching on the images that
were present.
On the non-vegetated area almost every configuration used was more than 90% of
the points correct, making the ATP Generator very precise to finding points in non-
65
forested areas, while on the forested area the percentages were from 0% to 40%. (See
Figure 4-20).
After conducting the experiment, there was not really a correlation between percent
of rights points and number of points found. The chosen options for ATP generation for
the assessment of LPS will be Search Area: 150 pixels, Coefficient: 0.9, Point Density:
10%, Initial Accuracy: 25%, this was the solution that gave the most auto tie points
possible without giving away accuracy. (See Figure 4-21).
Auto Tie Point Generation Assessment
After the tie point generation process, the auto tie summary dialog opens to display
the tie point generation for each image in the block file. After a quick review, a few of the
points where checked to ensure accuracy, adjusted in the case of possible adjustments,
and deleted if the tie points were completely wrong. Figure 4-22 shows that tie point
generation accuracy did not change with forward or side overlap percentages.
Figure 4-23 shows that there is a direct relation between tie points found and overlap
percentages, for higher percentage values more unique tie points are found
automatically. Also, a relation is shown on Figure 4-24 between tie points found and
total images in the block. Figure 4-25 shows the percent of the images that didn´t have
enough tie points, have clouds present on them, the percent of images with cloud that
were fulfilled were higher than 60% making this one of the reasons that this images
were bad tied together. Also the percentages of the images that had pitch or roll values
higher than 5 degrees are shown on Figure 4-25, being these considerable lower than
the cloud impact (30% to 75%).
66
Aerial Triangulation
After the tie point generation and check, and all images had 6 needed images to
proceed, the next step was performing an aerial triangulation, to adjust the final values
of the initial approximations to exterior orientation parameters provided by airborne GPS
and INS techniques. The technique used for projects with more than 2 images is the
bundle block adjustment, as space intersection and space resection techniques cannot
be performed. A bundle block adjustment as defined earlier is computed including
exterior orientation parameters, X, Y and Z coordinates of tie points and GCPs in some
cases, all the images are processed in one solution. A least square adjustment is used
to estimate the solution for the entire block while also minimizing and distributing the
error.
The options in the aerotriangulation dialog box, maximum iterations which I will use
the default 10, a convergence value in meters of 0.10 meters as proposed by the
developers, and the compute accuracy for unknowns will be used to calculate the
accuracy for parameters such as exterior orientation, ground coordinates and other
estimated by aerial triangulation, the images coordinates for reports in pixels as default.,
on the other option in the general tab, image point standard deviations will be left as
default (0.33 x 0.33 pixels), and GCP type and standard deviation was not used, as
GCPs were not used.
On the interior tab, the options were unchanged as any camera calibration was
properly done. For the standard deviation for exterior orientation, a same weighted
values for all images and the default values were used, 10 meters for X-Y-Z, and 5
degrees for pitch-roll-yaw, this values were left without change to try to establish a
better perception of software capacities, trying to do it as more default as possible.
67
For additional parameter model, there are several models that can be used in aerial
triangulation for the compensation of systematic image errors. The lens distortion model
was used according to developers recommendation that this is especially effective with
amateur digital cameras, this model has two additional parameters and is designed to
self-calibrate the lens distortion parameters automatically. These additional parameters
were used as weighted parameters, a small statistical weight value is assigned to each
parameter automatically by LPS.
As a blunder error detection, the time saving robust blunder checking which uses a
robust iterative weight function for gross error detection without computation of
individual redundancy for each observation.
After running the software a dialog opens showing if a triangulation convergence was
reached and total image unit-weight RMSE, if convergence was reached and the RMSE
had an acceptable value, less than 5 in cases that were possible the results can be
accepted and update the new exterior orientation parameters and proceed with the next
step of the process, ortho-rectification.
To improve the triangulation results or if the convergence could not be reached, a
review of the triangulation report and identification of the points with the most error,
usually with higher residuals, and delete them from triangulation calculations. Other
recommendation is to perform ATP generation using different options and
configurations, and even doing manual tie point, in order to have more points, so a
deletion of higher residual points can be made and the minimum points can still be have
in each image, continue an iteration of tie point generation and triangulation until a
correct adjustment can be done.
68
Figure 4-26 shows no relation at all between Root Mean Square Error (RMSE) and
overlap percentages, and considerable higher values than resulting EnsoMOSAIC data.
Ortho-rectification
DTM Extraction
The DTM Creation is the first step in the ortho-rectification process, and between the
different types of outputs, the one selected and more conventional file was a DEM, that
is a raster file that depicts elevation in dark and light pixels; dark pixels represent low
elevation areas, and bright represent high elevation areas, a single mosaic output file
was selected, and it will take each DTM created by each image pair in the block and will
be merged into one file, something similar to a mosaic. DEM Accuracy was left in
default value, 25 meters, this is the tolerance in the vertical units of the terrain to set
accuracy range for the predicted surface value of the area.
Ortho-Resampling
After the extraction of the DTM, the original images must be resampled
orthographically. For ortho-resampling DEM previously created in the DTM extraction
process was used. In this process the resampling method to be used was cubic
convolution, no rescaling used, the whole pixels where aligned in overlapping images,
and the resample was projected to UTM Zone 17 North projection, all other options kept
as default, as overlap threshold,
In this process all images can be done together, to ensure that all output ortho-
images have the same settings (for example, cell size, resampling method), set al.l of
the options in both the General and Advanced tab before adding additional images to
the process. Once these parameters have been set, they carry over to the additional
images added to the list.
69
Mosaic Creation
MosaicPro was the application within Erdas and LPS used to join georeferenced
images and form a larger image or a set of images. In this final step of the ortho-
rectification process, when using ortho resampled images, not much decisions should
be made, the only one is to set the overlap function, this one lets the user to set the
intersection type and overlap function for the mosaicked image, the options are: (a)
Overlay, the overlap area belongs to the image that is on top in the stacking order (b)
Average, the value of each pixel in the overlap area is replaced by the average of the
values of the corresponding pixels in the overlapping images (c) Minimum, the value of
each pixel in the overlap area is replaced by the lesser value of the corresponding
pixels in the overlapping images (d) Maximum, the value of each pixel in the overlap
area is replaced by the greater value of the corresponding pixels in the overlapping
images (e) Feather, the overlap area is replaced by a linear interpolation of the pixels in
the overlap. A pixel in the middle of the overlap area is 50% of each of the
corresponding pixels in the overlapping images. A pixel 1/10 of the overlap from an
edge would be 90% one image and 10% the other. This was the option that worked
better with the imagery used in this work.
70
Table 4-1. Sea Horse Key, Cedar Key Data Set Block Configuration.
Images/Strip
Overlap
Block Images Strips Average Min Max Height Average
Min
Max (Images) (meters) (meters) % (meters) % (meters) %
0 360 13 00 326 13 25.1 17 34 173.4 57.0 52% 44.9 83% 75.6 37%
11 117 5 23.4 17 34 174.7 53.4 56% 46.3 82% 69.7 42%
12 133 5 26.6 20 34 174.3 58.3 51% 44.9 83% 75.6 37%
13 133 5 26.6 21 33 171.5 58.4 51% 46.2 81% 71.9 39%
21 117 5 23.4 17 34 174.7 53.4 56% 46.3 82% 69.7 42%
22 243 9 27.0 20 34 172.9 57.3 52% 44.9 83% 75.6 37%
31 45 5 9.0 7 12 174.9 53.5 56% 46.0 82% 70.1 42%
32 39 5 7.8 6 9 174.3 53.5 55% 46.2 82% 69.4 42%
33 37 5 7.4 6 9 175.2 54.3 55% 45.7 83% 71.2 41%
34 42 5 8.4 5 12 173.8 59.3 50% 45.5 83% 96.8 19%
35 36 5 7.2 6 9 174.3 57.7 52% 44.7 84% 71.6 40%
36 39 5 7.8 6 10 174.4 57.4 52% 44.2 84% 71.0 41%
37 30 5 6.0 5 7 174.9 58.0 52% 45.2 83% 70.4 42%
38 42 5 8.4 6 11 170.9 60.1 49% 46.2 81% 76.7 35%
39 38 5 7.6 6 9 171.9 59.1 50% 45.3 82% 71.8 39%
310 36 5 7.2 6 9 171.9 55.8 53% 44.8 83% 68.4 42%
311 31 5 6.2 5 8 171.6 57.6 51% 42.3 86% 74.4 37%
71
Table 4-2. Total Processing Time Per Configuration
Configuration Blocks Minutes Seconds
0 1 18.92 1135
1 3 21.20 1272
2 2 20.63 1238
3 11 23.02 1381
Table 4-3. Processing Results
Time Seconds/ Connectivity Image Pairs Clust Total Area Macro Tot
Block Images Strips Seconds Minutes Image Nodes Edges Matched Tried (Units) (Pix x Pix) Tiles Ima
0 360 13 1230 20.50 3.42 00 326 13 1135 18.92 3.48 326 479 479 911 2 648188274 52 202
11 117 5 236 3.93 2.02 110 77 77 281 2 137414976 14 38
12 133 5 426 7.10 3.20 133 158 158 346 3 269828425 22 75
13 133 5 610 10.17 4.59 133 270 270 349 2 347254765 26 118
21 117 5 258 4.30 2.21 110 77 77 281 2 137570539 14 38
22 243 9 980 16.33 4.03 243 417 417 671 2 507867918 39 177
31 45 5 88 1.47 1.96 45 25 25 112 1 53252220 6 13
32 39 5 23 0.38 0.59 39 8 8 99 0
0
33 37 5 125 2.08 3.38 37 48 48 89 1 77604402 6 23
34 42 5 125 2.08 2.98 42 47 47 105 1 74924232 6 21
35 36 5 23 0.38 0.64 36 13 13 90 0
0
36 39 5 167 2.78 4.28 39 73 73 98 1 99815296 8 32
37 30 5 115 1.92 3.83 30 32 32 74 1 83812663 8 18
38 42 5 190 3.17 4.52 42 79 79 104 1 119244796 10 36
39 38 5 206 3.43 5.42 38 56 56 95 1 102287625 8 25
310 36 5 173 2.88 4.81 36 82 82 90 1 110358024 8 36
311 31 5 146 2.43 4.71 31 65 65 77 1 82119555 6 29
72
Table 4-4. Configuration and Options Used for ATP Experiment
Figure 4-1. Sea Horse Key Data Set.
Configuration Search Area
(Pixels) Coefficient Pnt Density (%) Initial
Accuracy (%)
1 96 0.8 10 25
2 96 0.9 10 25
3 96 0.9 10 10
4 96 0.9 50 10
5 96 0.9 50 25
6 96 0.9 100 10
7 96 0.9 100 25
8 150 0.9 10 10
9 150 0.9 100 10
10 150 0.9 100 25
11 150 0.9 10 25
73
Figure 4-2. Scaling error after mosaicking a big number of images.
Figure 4-3. Images Processed vs. Time Processed Chart. Cedar Key mission data using 2d3 Altimap.
0
50
100
150
200
250
300
350
400
0 200 400 600 800 1000 1200 1400
Imag
es
Time (seconds)
Images vs Time
74
Figure 4-4. Images Processed vs. Time Processed Chart for Block Configuration 3. Zoom in to Figure 4-3 to the Block Configuration 3, showing there is no clear relation between number of images to processing time, maybe due to the different features contained for every final mosaic (water, vegetation, others). Cedar Key data using 2d3 Altimap.
Figure 4-5. Chart showing the relations between average overlap to processing time. Every block configuration is represented with a different series to evaluate the true impact of overlap from block with similar number of images. Cedar Key Data using 2d3 Altimap.
25
30
35
40
45
50
0 50 100 150 200 250
Imag
es
Time (seconds)
Images vs Time (Blocks31-Blocks311)
53.000
54.000
55.000
56.000
57.000
58.000
59.000
60.000
61.000
0 200 400 600 800 1000 1200
Ave
rgae
Ove
rlap
(m
)
Time (Seconds)
Average Overlap vs Time
Configuration 1
Configuration 2
Configuration 3
Block00
75
Figure 4-6. Relation between final mosaic area with processing time per block. The chart is showing a good relation mosaic area-processing time. Cedar Key data using 2d3 Altimap.
Figure 4-7. Relationship between number of images utilized in the final mosaicked product and the processing time. Cedar Key data using 2d3 Altimap.
0
100000000
200000000
300000000
400000000
500000000
600000000
700000000
50 250 450 650 850 1050 1250
Mo
saic
Are
a (P
ixe
l x P
ixe
l)
Time (Seconds)
Area vs Processing Time
Config 1
Config 2
Config 3
Block00
0
50
100
150
200
0 200 400 600 800 1000 1200
Mo
saic
ked
Imag
es
Time (seconds)
Mosaicked Images vs Processing Time
Config 1
Config 2
Config 3
Block00
76
Figure 4-8. Relationship between Processing Time and Total Images in Block. Archer Fields data using 2d3 Altimap.
Figure 4-9. Relation between Image by Processing Minute and Forward Overlap. Datasets of 40% and 70% Side overlap. Archer Fields data using 2d3 Altimap.
0
50
100
150
200
250
300
350
400
0.00 5.00 10.00 15.00 20.00 25.00 30.00
Imag
es
Pro
cess
ed
Processing Time (Minutes)
Images vs Processing Time
Images vs Time
10.00
12.00
14.00
16.00
18.00
20.00
22.00
24.00
20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70% 75% 80%
Imag
e/P
roce
ssin
g Ti
me
(Im
age
s/M
inu
te)
Forward Overlap (%)
Forward Overlap vs Image/Processing Time
~40% Sidelap
~70% Sidelap
77
Figure 4-10. Scatter Plot Showing Relation Between Image Pairs Matched and Forward
Overlap. Archer Fields data using 2d3 Altimap.
Figure 4-11. Scatter Plot of Relation Between Images Mosaicked and Forward Overlap.
Archer Fields data using 2d3 Altimap.
60%
70%
80%
90%
100%
20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70% 75% 80%
Imag
e P
airs
Mat
che
d (%
)
Forward Overlap (%)
Forward Overlap vs Image Pairs Matched
~40% Sidelap
~70% Sidelap
60%
70%
80%
90%
20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70%
Imag
es
Mo
saic
ked
(%)
Forward Overlap (%)
Forward Overlap vs Images Mosaicked
~40% Sidelap
80%
84%
88%
92%
96%
20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70%
Imag
es
Mo
saic
ked
(%)
Forward Overlap (%)
Forward Overlap vs Images Mosaicked
~70% Sidelap
78
Figure 4-12. Scatter plot showing relation image percent with cloud and tree present.
Archer Fields data using 2d3 Altimap.
Figure 4-13. RMSE variation through measurements stages in Aerial Automatic
Triangulation. EnsoMOSAIC.
40%45%50%55%60%65%70%75%80%
60% 65% 70% 75% 80% 85% 90% 95%
Clo
ud
/Tre
e P
rese
nt
(%)
Mosaicked Images (%)
Mosaicked Images vs Cloud/Tree Present
Tree Present
Cloud Present
0
0.5
1
1.5
2
2.5
3
Initial (Level4)
Intermediate(Level 3)
Intermediate(Level 2)
Intermediate(Level 1)
Intermediate(Level 0)
Final (Level 0)
RM
SE (
Pix
els)
RMSE Summary through Aerial Automatic v7 Triangulation
Block 0
Block 1
Block 2
Block 3
Block 5
Block 6
Block 7
79
Figure 4-14. Scatter plot showing relation between overlap percentages with quantity of
tie points found on the initial stage. EnsoMOSAIC.
Figure 4-15. Scatter plot showing relation between total images in the block with
quantity of tie points found on the initial stage. EnsoMOSAIC.
0
500
1000
1500
2000
2500
20% 25% 30% 35% 40% 45% 50% 55% 60% 65% 70% 75% 80%Tie
Po
ints
Fo
un
d o
n In
itia
l Ru
n (P
oin
ts)
Forward Overlap (%)
Forward Overlap vs Tie Points Found
Side Overlap 40%
Side Overlap 70%
0
10
20
30
40
50
60
70
80
0 500 1000 1500 2000 2500
Tota
l Im
age
s (I
mag
es)
Tie Points Found (Points)
Tie Points Found vs Total Images
Blocks
80
Figure 4-16. Scatter plot showing relation between overlap percentages in the block
with accuracy of tie points found on the initial stage. EnsoMOSAIC.
.
Figure 4-17. Cloud presence and high pitch-roll values impact on tie point search.
EnsoMOSAIC.
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
20% 30% 40% 50% 60% 70% 80%
Tie
Po
int
Acc
ura
cy (%
)
Forward Overlap (%)
Forward Overlap vs Tie Point Accuracy
Side Overlap 40%
Side Overlap 70%
0%
20%
40%
60%
80%
100%
Block0
Block1
Block2
Block3
Block4
Block5
Block6
Block7
Block8
Percent Not Used w/Cloud 0% 60% 100% 100% 67% 100% 60% 27% 100%
Percent Not Used w/ High Titl Angles 60% 40% 67% 40% 78% 44% 60% 73% 57%
Pe
rce
nt
of
No
t U
Sed
Not Used with Clouds/High Tilts
81
Figure 4-18. Scatter plot showing relation between overlap percentages in the block
with Root Mean Square Errors on the Aerial Triangulation process. EnsoMOSAIC.
Figure 4-19. Scatter plot showing relation Root Mean Square Errors on the Aerial
Triangulation process and percent of images used for mosaicking. EnsoMOSAIC.
0.60.8
11.21.41.61.8
22.22.42.62.8
3
20% 30% 40% 50% 60% 70% 80%
RM
SE (
Pix
els
)
Forward Overlap (%)
Forward Overlap vs RMSE
Side Overlap 40%
Side Overlap 70%
50%
55%
60%
65%
70%
75%
80%
85%
90%
95%
100%
0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
Per
cen
t Im
age
s U
sed
RMSE (Pixels)
Final RMSE vs Percent Images Used
Blocks
82
Figure 4-20. ATP Experiment Accuracy per Configuration. Erdas LPS.
Figure 4-21. Scatter Plot Showing Percent Right vs. ATP Found. Erdas LPS.
0.00%
20.00%
40.00%
60.00%
80.00%
100.00%
1 2 3 4 5 6 7 8 9 10 11
ATP Experiment Configurations Accuracies
No-Vegetaton Right (%) Vegetation Right (%) Total Right
2000
2200
2400
2600
2800
3000
3200
3400
40% 45% 50% 55% 60% 65% 70% 75%
Po
ints
Fo
un
d
Percent Right (%)
Percent Right-Points Found (Point Density Fixed)
Point Density 10%
Point Density 50%
Point Density 100%
Options configuration used (No. 11)
83
Figure 4-22. Scatter plot showing relation between tie point accuracy and overlap
percentages. Erdas LPS.
Figure 4-23. Scatter plot showing relation between overlap percentages with quantity of
tie points found on the initial stage. Erdas LPS.
40%45%50%55%60%65%70%75%80%85%90%95%
100%
20% 30% 40% 50% 60% 70% 80%
Tie
Po
int
Acc
ura
cy (%
)
Forward Overlap (%)
Forward Overlap vs Tie Point Accuracy
Side Overlap 40%
Side Overlap 70%
0
100
200
300
400
500
600
700
800
900
1000
20% 30% 40% 50% 60% 70% 80%
Tie
Po
ints
Fo
un
d (P
oin
ts)
Forward Overlap (%)
Forward Overlap vs Tie Points Found
Side Overlap 40%
Side Overlap 70%
84
Figure 4-24. Scatter plot showing relation between total images in the block with
quantity of tie points found on the initial stage. Erdas LPS.
Figure 4-25. Cloud presence and high pitch-roll values impact on tie point search.
Erdas LPS.
0
200
400
600
800
1000
0 10 20 30 40 50 60 70 80
Po
ints
Fo
un
d (P
oin
ts)
Images (Images)
Images vs Tie Points
Blocks
0%
20%
40%
60%
80%
100%
Block0
Block1
Block2
Block3
Block4
Block5
Block6
Block7
Block8
Percent Not Used w/Clouds 0% 100% 67% 100% 60% 100% 75% 100% 100%
Percent Not Used w/High TitleAngles
50% 60% 67% 67% 80% 75% 75% 50% 33%
Per
cen
t o
ut
of
No
t U
Sed
Not Used with Clouds/High Tilt
85
Figure 4-26. Relation between overlap percentages and Root Mean Square Error
(RMSE) in aerotriangulation. Erdas LPS.
0
5
10
15
20
25
0% 10% 20% 30% 40% 50% 60% 70% 80%
Ae
rial
Tri
angu
lati
on
RM
SE (
pix
els
)
Forward Overlap (%)
Forward Overlap vs RMSE
Side Overlap 40%
Side Overlap 70%
Blocks not converged
86
CHAPTER 5 CONCLUSIONS
Comparison
2d3 Altimap created visually pleasant mosaics in all configurations done with
smooth seam-lines, great for GIS and Remote Sensing image processing applications.
The mosaicking in the trees were good also with minimum errors. The errors found in
these mosaics were located in the edges of the area of study, where no overlap
occurred. 2d3 Altimap also applied auto level corrections and color balancing to the
images, correcting shadow effect by clouds. Figure 5-1 shows a mosaic created by 2d3
Altimap and the zoom to one matching error, where the street was discontinued.
EnsoMOSAIC created visually pleasant mosaics as well. The histogram matching
between images created the smoother seam-lines in the mosaics by EnsoMOSAIC.
These outputs looked pretty similar to the final result of 2d3 Altimap. These resulting
mosaics can be used for GIS and Remote Sensing image processing applications. Sun
corrections can be done automatically by the software if UTC date and time where
inputted on the first step of the post-processing. Figure 5-2 shows the resulting mosaic
with the histogram matching and bilinear interpolation resampling options.
Erdas LPS created the least pleasant mosaics of the three software programs
evaluated, there was not a proper automatic seam-line creation, and no sun correction
or color balancing where applied, for this the image has to go through more radiometric
manipulation. A mosaic created by LPS is shown by Figure 5-3.
EnsoMOSAIC was capable than Erdas LPS in finding more automatic tie points on
the first interaction of ATP process done with similar options for both software, as
shown in Figure 5-4. This is true for every block configuration used. Figure 5-5 shows
87
that even though EnsoMOSAIC found more points also in general LPS had more tie
point found accuracy with more of the blocks assessed had more than 60% of tie points
right up to 90%.
Despite the fact that LPS had more tie points accuracy, the algorithm used by
EnsoMOSAIC results with less Root Mean Square Error (RMSE). Figure 5-6 shows the
comparison in RMSE(Pixels) between both software.
2d3 Altimap can process mosaics within hours of data acquisition while the two
other software can take more than five days, working 8-plus hours per day, for standard
missions of 300 to 400 images.
Conclusions
2d3 Altimap is a good 2D georeferenced mosaic solution for UAS imagery, not as
its price suggests (US$150.00). The mosaics created by Altimap are very well
georeferenced, even if it does not contemplate exterior orientation parameters (pitch,
roll and yaw angles). Between the packages that use exterior orientation parameter, the
better equipped for UAS imagery mosaicking is EnsoMOSAIC resulting in better
resampled ortho-mosaics and lower RMSE adjustments. The DEM generated by both
EnsoMOSAIC and LPS were not very reliable, compared to existing DEM.
My recommendations are that UFUASRG and ACOE buy 2d3 Altimap solution for
fast mosaic applications with good accuracy. I also recommend to use EnsoMOSAIC for
missions using GCPs for better accuracies, and for flights with high winds that make the
aircraft flights rougher.
Also, for better mosaics a camera calibration protocol has to be created for the
current payload, where the calibration procedure would be done periodically, every
certain amount of flights. Further a payload with a better camera would help in the
88
processing, like an industrial digital camera that the UFUASRG is trying to implement.
For better positional accuracies, and perhaps better image matching in processing, a
payload with differential or RTK GPS aboard would improve these positional accuracy
of the system.
The three software assessed were able to process at less than standard overlap
percentages (60% forward, 35% side), but the flight planning has to be designed for at
least these values for better processing.
Recommendations
In future research about this subject, a positional assessment of the software has
to be done. High resolution aerial or satellite images updated to any particular mission
to be done are rare, for this kind of project concrete plates, monuments or markings like
the ones used for GCPs have to be included in the area of study. This control points
have to be measured with high accuracy surveying techniques, like RTK GPS, to
achieve accurate coordinates, and this coordinates to be assessed with the final
mosaics of the different software.
Using high accuracy surveying techniques a DEM have to be created and be
compared with the DEM created by the software pixel by pixel to achieve a good
elevation measurement capability assessment of the software.
Also, the results should be compared using calibrated camera parameters and
without them, in order to get the impact of camera calibration procedures in post-
processing for UAS photogrammetry.
89
Figure 5-1. Resulting mosaic using 2d3 Altimap. (Left) Mosaic of Block 0 done with 2d3
Altimap (Right) zoom to west zone of mosaic showing smoothness of seam lines and matching errors.
90
Figure 5-2. Resulting mosaic using EnsoMOSAIC. (Left) Mosaic of Block 0 done with
EnsoMOSAIC (Right) zoom to west zone of mosaic showing smoothness of seam lines (histogram matching option).
Figure 5-3. Resulting mosaic using Erdas LPS. (Left) Mosaic of Block 6 done with
Erdas LPS (Right) zoom to west zone of mosaic showing smoothness of seam lines (average option).
91
Figure 5-4. Comparison between EnsoMOSAIC and Erdas LPS in automatic tie point
found.
Figure 5-5. Comparison between EnsoMOSAIC and Erdas LPS in automatic tie point
accuracy.
0
500
1000
1500
2000
2500
Block 0 Block 1 Block 2 Block 3 Block 4 Block 5 Block 6 Block 7 Block 8
Tie
Po
ints
Fo
un
d (Q
uan
tity
)
Block Configurations
Tie Points Found Comparison
EnsoMOSAIC
Erdas LPS
0%
20%
40%
60%
80%
100%
Block 0 Block 1 Block 2 Block 3 Block 4 Block 5 Block 6 Block 7 Block 8
Tie
Po
int
Acc
ura
cy (%
)
Block Configurations
Tie Points Accuracy Comparison
EnsoMOSAIC
Erdas LPS
92
Figure 5-6. Comparison between EnsoMOSAIC and Erdas LPS in aerial triangulation
error.
0
5
10
15
20
25
Block0
Block1
Block2
Block3
Block4
Block5
Block6
Block7
Block8
Tria
ngu
lati
on
RM
SE (
Pix
els
)
Block Configurations
Aerial Triangulation Error Comparison
EnsoMOSAIC
Erdas LPS
93
APPENDIX IMAGE EXIF DATA READERS
For the 2d3 Altimap, one important part of the processing steps is to write the
EXIF data to the collected imagery from the UAV. There are great online tools that can
be used to verify the data is written correctly to the images. The following are two of the
tools that give more information, including geotags of the image, camera info, and other
parameters on the moment of the capture: http://regex.info/exif.cgi/exif.cgi and
http://www.exifdata.com . (See Figures A-1 and A-2 for screen captures of examples of
both pages).
Figure A-1. Jeffrey´s Exif Viewer, showing the location of an Archer Field Mission
Image using the EXIF info written to it. Source (http://regex.info/exif.cgi/exif.cgi).
94
Figure A-2. Example of all the data contained in EXIF format written in an image. Source: (www.exifdata.com).
95
LIST OF REFERENCES
Aber, J.S., I. Marzolff and J.B. Ries, 2010. Small-Format Aerial Photography, Principles, Techniques and Geoscience Applications, First Edition, Elsevier, Amsterdam, The Netherlands, 266 p.
Ackermann, F., 1983. High precision digital image correlation Paper presented at 39th Photogrammetric Week, Institute of Photogrammetry, University of Stuttgart, pp. 231-243.
Afek, Y. and A. Brand, 1998. Mosaicking of Orthorectified Aerial Images, Photogrammetric Engineering & Remote Sensing, 64 (2): 115-125.
American Society of Photogrammetry,1980. Photogrammetric Engineering and Remote Sensing, 46 (10): 1249.
Burgess, M.A., P.G. Ifju, H.F. Percival, J.H. Perry, and S.E. Smith, 2009. Development of a survey-grade sUAS for natural resource assessment and monitoring, Slide presentation, Florida Cooperative Fish and Wildlife Research Unit Annual Coordinating Committee Meeting, Gainesville, FL. 32611.
Buyuksalih, G. and Z. Li, 2003. Practical Experiences with Automatic Aerial Triangulation Using Different Software Packages, Photogrammetric Record, 18 (102): 131-155.
Campbell, J.B., 2011 Introduction to Remote Sensing, Fifth Edition, Guilford Press, New York, 684 p.
Digicamhelp. “What is ISO?” Digital camera help for beginners and beyond, Digicam Website. URL: http://www.digicamhelp.com/camera-features/advanced-settings/iso/ (last date accessed: 18 July 2012).
Eisenbeiss, H., 2009. UAV Photogrammetry, Doctorate Dissertation, Institute of Geodesy and Photogrammetry, ETH Zurich. Zurich, 203 p.
Erdas Inc., 2010. LPS Project Manager User´s Guide, United States of America, 421 p.
Everaerts, J., 2008. The Use of Unmanned Aerial Vehicles (UAVS) For Remote Sensing and Mapping. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. XXXVII ( B1): 1187-1191
Exif Data Viewer. “What is EXIF Data?” EXIFDATA Website, URL: http://www.Exifdata.com (last accessed: 25 July 2012).
Förstner, W., and E. Gülch, 1987. A fast operator for detection and precise location of distinct points, corners and centers of circular features, Paper presented at the Intercommission Conference on Fast Processing of Photogrammetric Data, Interlaken, Switzerland, pp. 281-305.
96
Ghosh, S.K., 2005. Fundamentals of Computational Photogrammetry, Concept Publishing Company, New Delhi, India, 254 p.
Granshaw, S. I., 1980. Bundle Adjustment Methods In Engineering Photogrammetry. The Photogrammetric Record, 10 (56): 181–207.
Grün, A. and E. P. Baltsavias, 1988. Geometrically constrained multiphoto matching, Photogrammetric Engineering and Remote Sensing 54 (5): 633-641.
Haarbrink, R.B. and E. Koers, 2006. Helicopter UAV for Photogrammetry and Rapid Response. The International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, Antwerp, Belgium, XXXVI(1-W44): 4 p.
Heipke, C., 1997. Automation of interior, relative, and absolute orientation, ISPRS Journal of Photogrammetry and Remote Sensing, 52(1): 1-19.
Helava, U. V., 1988. Object space least square correlation, International Archives of Photogrammetry and Remote Sensing, 27 (B3): 321-331.
Herwitz, S.R., L.F. Johnson, S.E. Dunagan, R.G. Higgins, D.V. Sullivan, J. Zheng, B.M. Lobitz, J.G. Leung, B.A. Gallmeyer, M. Aoyagi, R.E. Slye and J.A. Brass, 2004. Imaging from an unmanned aerial vehicle: agricultural surveillance and decision support. Computers and Electronics in Agriculture 44 (1): 49-61.
Ionescu, D., 2010. Geolocation 101: How It Works, The Apps, and Your Privacy, PCWorld, URL: http://www.pcworld.com/article/192803/geolocation_101_how_it_works_the_apps_and_your_privacy.html (last accessed: 23 July 2012)
Jeffrey´s EXIF. Jeffrey´s EXIF Viewer, URL: http://regex.info/exif.cgi/exif.cgi (last accessed 13 October 2012).
Kasser, M. and Y Egels, 2002. Digital Photogrammetry, First Edition, Taylor & Francis, New York, 351 p.
Kennedy, R. and W.Cohen, 2003. Automated designation of tie-points for image-to-image coregistration, International Journal of Remote Sensing, 24(17): 3467-3490.
Konecny, G., 1994. New Trends in Technology, and their Application: Photogrammetry and Remote Sensing—From Analog to Digital. Thirteenth United Nations Regional Cartographic Conference for Asia and the Pacific, Beijing, China, 6 p.
Krzystek, P., T. Heuchel, U. Hirt and F. Petran, 1996. An Integral Approach to Automatic Aerial Triangulation and Automatic DEM Generation, International Archives of Photogrammetry and Remote Sensing, 31 (B3): 405-414.
97
Laliberte, Andrea, 2012. Mapping Rangelands With Unmanned Aircraft, Aerial Mapping Spring 2012, Professional Surveyor Magazine, URL: http://www.profsurv.com/magazine/article.aspx?i=71086 (last accessed: 23 October 2012).
Lillesand, T., R. Kiefer and J. Chipman, 2008. Remote Sensing And Image Interpretation. Sixth Edition. Wiley and Sons. Madison, Winsconsin, 763 p.
Liu, X., P. Chen, X. Tong, S. Liu, S. Liu, Z. Hong, L. Li and K. Luan, 2012. UAV-based low-altitude aerial photogrammetric application in mine areas measurement, Earth Observation and Remote Sensing Applications (EORSA), 2012 Second International Workshop: 240-242.
Matthews, N. A., T.A. Noble and B.H. Breithaupt, 2012. Dinosaur tracks to dam faces: A new method for collecting three-dimensional data, Geological Society of America Abstracts 36(5): 348 p.
Morgado, A., and I. Dowman, 1997. A procedure for automatic absolute orientation using aerial photographs and a map, ISPRS Journal of Photogrammetry & Remote Sensing 52(4): 169-182.
MosaicMill Oy, 2009. EnsoMOSAIC Image Processing User´s Guide, Finland, 106 p.
Norut (Northern Research Institute). UAV Applications, Nortu Website, URL: http://uas.norut.no/UAV_Remote_Sensing/Applications.html (last accessed: 14 October 2012).
Perry, J., 2009. A Synthesized Directly Georeferenced Remote Sensing Technique for Small Unmanned Aerial Vehicles, Masters Thesis, School of Forest Resources and Conservation, University of Florida, Gainesvile, Florida, 154 p.
Qingyuan, L., L. Wenyang, Z. Lei, W. Jingling and L. You, 2011. A New Approach to Fast Mosaic UAV Images, International Archives of the Photogrammetry XXXVIII (1/C22): 6 p.
Rango, A. A. Laliberte, C. Steele, J.E. Herrick, B. Bestelmeyer, T. Schmugge, A. Roanhorse and V. Jenkins, 2006. Using unmanned vehicles for rangelands: Current applications and future potentials. Environmental Practice, 8 (3): 159-169.
Réstas, A., 2006. The regulation Unmanned Aerial Vehicle of the Szendro Fire Department supporting fighting against forest fires, Forest Ecology and Management, 234S (S233): 234-240.
Sawhney, H.S. and R. Kumar, 1999. True multi-image alignment and its application to mosaicking and lens distortion correction, Pattern Analysis and Machine Intelligence, IEEE Transactions, 21 (3): 235-243.
98
Schenk, T., 1997. Towards automatic aerial triangulation, ISPRS Journal of Photogrammetry & Remote Sensing, 52 (3): 110-121.
Schenk, T., 2004. From point-based to feature-based aerialtriangulation, ISPRS Journal of Photogrammetry and Remote Sensing, 58 (5-6): 315-329.
Shragai, Z, S. Barnea and A. Even-Paz, 2012. Novel Matching Method for Handling Non-Standard Imagery: Automatic Tie-point Extraction. GIM International, USA, pp 31-35.
Tang, L. J. Braun and R. Debitsch, 1997. Automatic aerotriangulation — concept, realization and results, ISPRS Journal of Photogrammetry and Remote Sensing, 52 (3): 122-131.
Turner, D. A. Lucier and C. Watson, 2012. An Automated Technique for Generating Geo-rectified Mosaics from Ultra-High Resolution Unmanned Aerial Vehicle (UAV) Imagery, Based on Structure from Motion (SfM) Point Clouds, Remote Sensing Journal, 4(5): 1392-1410.
Vosselman, G., and N. Haala, 1992. Recognition of topographic control points through relational mapping, Journal of Photogrammetry and Remote Sensing 60 (6): 170-176.
Vosselman, G., 1995. Applications of tree search methods in digital photogrammetry, International Society of Photogrammetry and Remote Sensing (ISPRS) Journal of Photogrammetry and Remote Sensing, 50 (4): 29-37.
Wang, Y., 1998. Principles and applications of structural image matching, International Society of Photogrammetry and Remote Sensing (ISPRS) Journal of Photogrammetry and Remote Sensing 53 (3): 154-165.
Watts, A., F. Percival, S. Bowman,D. Coffin,P. Ifju, A. Abd-Elrahman, B. Dewitt, Y. Kaddoura, A. Mohamed, J. Lane, K. Lee, L. Pearsltine, J. Perry and S.E. Smith,2010. Scientific Applications of UAS: Biological Research and Natural Resource Monitoring, University of Florida, Gainesville, FL. 32611.
Wilkinson, B.E., 2007. The Design of Georeferencing Techniques for an Unmanned Autonomous Aerial Vehicle for Use with Wildlife Inventory Surveys: A Case Study of the National Bison Range, Montana. Master’s Thesis, School of Forest Resources and Conservation, University of Florida, 115 p.
Wolf, P.R. and B.A.Dewitt, 2010. Elements of photogrammetry with applications in GIS, Third Edition, McGraw Hill, Boston, Massachusetts, 608 p.
Zebedin, L., A. Klaus, B. Gruber-Geymayer and K. Karner, 2006. Towards 3D map generation from digital aerial images, ISPRS Journal of Photogrammetry & Remote Sensing 60 (6): 413–427.
99
BIOGRAPHICAL SKETCH
Héctor Yamil Rodríguez Asilis decided at a really young age that he was going to
be a Civil Engineer, around ten years old, and everything that happened after that in
middle and high school just confirmed his desire. His father is a Civil Engineer and
Surveyor and his mother is a Civil Engineer as well, this made an environment for Yamil
at very young ages around plans, maps and construction to a point that a client was at
his dad office when he was 3 years old, and this baby boy entered the office took a
project plan and explain what the legend meant on the plans, so starting at that age he
knew at least how to read plans and maps. He is an avid golfer, playing almost to
scratch golfer, and being on the Junior Dominican National Team twice and winning his
Match Play Club Championship on his first try, and the younger one to date to
accomplish it.
Yamil majored in Civil Engineering and was one of the few that graduated in the 4
years the program proposes, and with “Cum Laude” honors for having a GPA of 3.3 out
of 4.0. One curious thing about Yamil, is that even though Civil Engineering has a
relatively extensive field, he was always comfortable in every field he studied and
worked, making this a tough decision on what area to make his specialization in
graduate school. He co-opted for the biggest engineering company in the island,
working as a field engineer. After graduation he went back to his beloved small town of
Puerto Plata to work on the family owned engineering company with his parents,
starting right out of the gate in the construction biggest commercial plaza ever on town
as a field engineer, and working simultaneously on the geomatics division of the same
company, where he later also did different certificates in boundary surveying and RTK
GPS and Differential GPS Surveying, and after a good look at the market, and already
100
fascinated with the geomatics he decided to apply for graduate studies in the area of
geomatics, ultimately deciding to go to University of Florida.
After a few semesters in the program he was hired on the UAV Project on the
post-processing team, making this a great experience and helped him learn a lot about
this new technology, the one that he will be venturing to take back to his home country
as a business where he thinks there is a lot of applications with this in the DR. He is
planning to graduate with a Masters of Science in Forest Resources and Conservation
with a Concentration in Geomatics in December 2012. His time in Gainesville and
University of Florida was great, he learned a lot of things, and met a lot of people who at
the absence of his family, who were back in the DR, became his adoptive family in the
good and the bad times. Go Florida Gators!!
